DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
This action is in response to amendments and remarks filed on 04/28/2026. Claims 1 and 4-15 are pending. Claims 2-3 have been cancelled. Claims 1 and 4-5 have been amended. Claims 6-15 have been added. The specification has been amended. The objection to the specification and the 35 U.S.C. 112 rejection to the claims have been withdrawn in light of the instant amendments. This action is made final, as necessitated by amendment.
Response to Arguments
Applicant's arguments filed 04/28/2026 have been fully considered but they are not persuasive. Applicant argues that Jun, Hong, and DiCarlo do not disclose the technology to set the acceleration threshold value appropriately.
Regarding Jun, Applicant argues that Jun does not teach learning the G-value and appropriately setting the G-value threshold, and therefore it does not teach claim 1. Applicant agrees that Jun alone does not teach claim 1. Further details on how Jun is used to teach the amended claim 1 can be found below under Claim Rejections.
Regarding Hong, Applicant argues that Hong teaches estimating the cause of the impact instead of learning an acceleration threshold value to determine whether an accident has occurred, and therefore does not teach claim 1. Hong teaches that a machine learning model uses acceleration data in order to estimate if the impact is caused by a closing of a door or if the impact is caused by something else (par. 16, “In exemplary embodiments, the learning process of the machine learning model used to estimate that an impact caused by opening or closing a door or an impact other than door opening and closing occurs in the parked vehicle, i) the parking An impact test caused by opening or closing a vehicle door and an impact test other than door opening and closing are performed to collect acceleration data and air pressure data of the vehicle before and after the occurrence of each impact, and record the type of each impact. doing; and ii) a characteristic element obtained from the collected acceleration data and air pressure data, which is used as an input of the machine learning model, and an impact corresponding to each of the acceleration data and each air pressure data among the types of recorded impacts It may include a step of setting the type as an output of the machine learning model, whereby the machine learning model can learn appropriate parameters for the input and the corresponding output”). Although it does not explicitly teach that the machine learning model is determining a threshold, in order to categorize the impact as either an impact caused by opening or closing the door or an impact caused by something else, the machine learning model would need to determine some sort of categorization threshold to differentiate between the two type of impacts. Since the model uses a characteristic element from the acceleration data, this threshold would need to be related to the acceleration data. Further details on how Hong is used to teach the amended claim 1 can be found below under Claim Rejections.
Regarding DiCarlo, Applicant argues that DiCarlo does not teach the technology in which the video is recorded when the timing of the door of the vehicle changing from the open to the closed state and the timing of accident occurrence are the same. However, nowhere in claim 1 does it state that this is required. Applicant agrees that DiCarlo alone does not teach claim 1. Further details on how DiCarlo is used to teach the amended claim 1 can be found below under Claim Rejections.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1 and 4-5 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jun (US 20210012589) in view of DiCarlo (US 20200406727) and Hong (KR 20230061711).
Regarding claim 1, Jun teaches a drive recorder device (Fig. 1, vehicle image recording apparatus 100), comprising: a memory (storage 120) and a processor (processor 140), the processor configured to:
determine whether an impact equal to or greater than an acceleration threshold value stored in the memory (par. 43, “The processor 140 may determine whether an impact of a predetermined reference value or more that is applied to the vehicle corresponds to an impact caused by an accident”) is detected by an acceleration sensor (Fig. 1 and par. 50, “sensing device 200 may include an impact sensor”);
start video recording (Fig. 3, S205 perform impact image recording) in the memory when the impact equal to or greater than the acceleration threshold value stored in the memory is detected by the acceleration sensor (par. 43, "The processor 140 may determine whether an impact of a predetermined reference value or more that is applied to the vehicle corresponds to an impact caused by an accident") while an ignition power supply of a vehicle is turned off (Fig. 2 S100, vehicle enters parking recording mode);
determine whether a door of the vehicle changes from an open state to a closed state (Fig. 2 S300, monitor door communication signal);
wherein (abstract, “a processor that determines whether an impact of a reference value or more applied to the vehicle is caused by an accident”—the impact is used to determine if an accident has occurred, so obviously the reference value will need to be a threshold value that would indicate an accident),
Jun fails to teach upon determining that the door changes from the open state to the closed state, perform learning of the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point, at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point, wherein the processor is configured to perform the learning of the acceleration threshold value so that the acceleration threshold value is greater than the impact detected by the acceleration sensor when the door changes from the open state to the closed state, and less than or equal to the impact detected by the acceleration sensor when an accident occurs, and the processor is configured to not use the impact detected by the acceleration sensor and associated with occurrence of the accident during the learning period in the learning of the acceleration threshold value.
Jun instead only teaches checking whether the impact is a predetermined reference value or more, and does not teach how the reference value is determined. The reference value is also not used to differentiate between a door closing event and a collision event. Instead, it uses the door sensor to do so (Fig. 3, S300). However, using a threshold value to differentiate between a door closing event and a collision event is already well-known in the field.
DiCarlo teaches determine whether an impact equal to or greater than an acceleration threshold value stored in the memory (abstract, “determining an event signature based on the sensor data”; par. 46, “In some examples, the event signature may be characterized by the amplitude of initial impulse peak 514 exceeding a threshold value”),
(par. 59, “In some examples, a machine learning model may be applied to the event signature methods independently or in a combined manner. The machine learning model may be developed by providing training sets of known gyroscope and/or sound level data associated with door closing events to the model”),
wherein the processor is configured to perform the learning of the acceleration threshold value so that the acceleration threshold value is greater than the impact detected by the acceleration sensor when the door changes from the open state to the closed state, and less than or equal to the impact detected by the acceleration sensor when an accident occurs (par. 59, “For example, if a vehicle receives an impact not due to a door closing (e.g. from an external object/person impacting the vehicle or from a passing truck creating an air pressure wave) an event signature from a gyroscope alone may falsely indicate a door closing event. A vehicle impact not due to a door closing event may generate a sound level signature distinguishable from a door closing sound level signature while the gyroscope data signature may not be able to distinguish between the door closing event and an impact due to another source. By correlating the gyroscope and sound level signatures a more robust (e.g. reducing false positive detection and/or false negative detection) event signature detection method is created”—DiCarlo teaches using both gyroscope data and sound data to differentiate between a door closing event and a collision event. Although not explicitly taught, the machine learning model would need to learn some kind of acceleration threshold that correlates with a sound level signature such that the acceleration teaches whether the event was a door closing event. It would be obvious to use a threshold that lies between a non-collision and a collision, and such thresholds are well-known in the field (examples include Hyuk (KR 20110005994), Agata US 20220410675)),
and the processor is configured to not use the impact detected by the acceleration sensor and associated with occurrence of the accident during the learning period in the learning of the acceleration threshold value (par. 59, “The machine learning model may be developed by providing training sets of known gyroscope and/or sound level data associated with door closing events to the model”—DiCarlo teaches just using door closing events to develop the learning model).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Jun to incorporate the teachings of DiCarlo in order to detect vehicle door closing events and to “increase the accuracy of detecting door closing events while reducing the occurrence of false positive and false negative door closing detection” (par. 51). Learning how to identify a door slam would also include learning an expected acceleration range for a door slam. Detecting an abnormal value (for example, a non-door slam acceleration value) to determine a possible crash is well-known in the art (examples include Hyuk (KR 20110005994), Agata US 20220410675)), and would have been an obvious modification.
DiCarlo fails to teach upon determining that the door changes from the open state to the closed state, perform learning of the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point, at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point.
DiCarlo instead only teaches using “training sets of known gyroscope and/or sound level data associated with door closing events to the model” (par. 59). This training set data would presumably have been gathered from actual door closing data, which would have been gathered during a learning period between a first time point, at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point.
However, Hong teaches upon determining that the door changes from the open state to the closed state, perform learning of the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point, at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point (par. 16, “the learning process of the machine learning model used to estimate that an impact caused by opening or closing a door or an impact other than door opening and closing occurs in the parked vehicle, i) the parking An impact test caused by opening or closing a vehicle door and an impact test other than door opening and closing are performed to collect acceleration data and air pressure data of the vehicle before and after the occurrence of each impact, and record the type of each impact”).
Hong teaches that these impact tests are done while a door is opened and closed. The machine learning model would need to be taught that an impact is done by a door closing event, therefore it would be determined whether or not a door is changing from an open to a closed state. Although it does not explicitly teach that the machine learning model is determining a threshold, in order to categorize the impact as either an impact caused by opening or closing the door or an impact caused by something else, the machine learning model would need to determine some sort of categorization threshold to differentiate between the two type of impacts. Since the model uses a characteristic element from the acceleration data, this threshold would need to be related to the acceleration data. It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Jun in view of DiCarlo to incorporate the teachings of Hong to include the learning process during an opening and closing of the door. Repeatedly learning can improve the machine learning model accuracy in determining a door closing event (par. 35).
Regarding claim 4, Jun teaches a video recording method comprising:
determining whether an impact equal to or greater than an acceleration threshold value stored in a memory (par. 43, “The processor 140 may determine whether an impact of a predetermined reference value or more that is applied to the vehicle corresponds to an impact caused by an accident”) is detected by an acceleration sensor (Fig. 1 and par. 50, “sensing device 200 may include an impact sensor”);
starting video recording (Fig. 3, S205 perform impact image recording) in the memory when the impact equal or greater than the acceleration threshold value stored in the memory is detected by the acceleration sensor (par. 43, "The processor 140 may determine whether an impact of a predetermined reference value or more that is applied to the vehicle corresponds to an impact caused by an accident") while an ignition power supply of a vehicle is turned off (Fig. 2 S100, vehicle enters parking recording mode);
determining whether a door of the vehicle changes from an open state to a closed state (Fig. 2 S300, monitor door communication signal);
wherein (abstract, “a processor that determines whether an impact of a reference value or more applied to the vehicle is caused by an accident”—the impact is used to determine if an accident has occurred, so obviously the reference value will need to be a threshold value that would indicate an accident),
Jun fails to teach upon determining that the door changes from the open state to the closed state, learning the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point, wherein learning the acceleration threshold value so that the acceleration threshold value is greater than the impact detected by the acceleration sensor when the door changes from the open state to the closed state, and less than or equal to the impact detected by the acceleration sensor when an accident occurs, and not using the impact detected by the acceleration sensor and associated with occurrence of the accident during the learning period in learning of the acceleration threshold value.
Jun instead only teaches checking whether the impact is a predetermined reference value or more, and does not teach how the reference value is determined. The reference value is also not used to differentiate between a door closing event and a collision event. Instead, it uses the door sensor to do so (Fig. 3, S300). However, using a threshold value to differentiate between a door closing event and a collision event is already well-known in the field.
DiCarlo teaches determining whether an impact equal to or greater than an acceleration threshold value stored in a memory (abstract, “determining an event signature based on the sensor data”; par. 46, “In some examples, the event signature may be characterized by the amplitude of initial impulse peak 514 exceeding a threshold value”),
point, which is a predetermined time before the first time point (par. 59, “In some examples, a machine learning model may be applied to the event signature methods independently or in a combined manner. The machine learning model may be developed by providing training sets of known gyroscope and/or sound level data associated with door closing events to the model”),
wherein learning the acceleration threshold value so that the acceleration threshold value is greater than the impact detected by the acceleration sensor when the door changes from the open state to the closed state, and less than or equal to the impact detected by the acceleration sensor when an accident occurs (par. 59, “For example, if a vehicle receives an impact not due to a door closing (e.g. from an external object/person impacting the vehicle or from a passing truck creating an air pressure wave) an event signature from a gyroscope alone may falsely indicate a door closing event. A vehicle impact not due to a door closing event may generate a sound level signature distinguishable from a door closing sound level signature while the gyroscope data signature may not be able to distinguish between the door closing event and an impact due to another source. By correlating the gyroscope and sound level signatures a more robust (e.g. reducing false positive detection and/or false negative detection) event signature detection method is created”—DiCarlo teaches using both gyroscope data and sound data to differentiate between a door closing event and a collision event. Although not explicitly taught, the machine learning model would need to learn some kind of acceleration threshold that correlates with a sound level signature such that the acceleration teaches whether the event was a door closing event. It would be obvious to use a threshold that lies between a non-collision and a collision, and such thresholds are well-known in the field (examples include Hyuk (KR 20110005994), Agata US 20220410675)),
and not using the impact detected by the acceleration sensor and associated with occurrence of the accident during the learning period in learning of the acceleration threshold value (par. 59, “The machine learning model may be developed by providing training sets of known gyroscope and/or sound level data associated with door closing events to the model”—DiCarlo teaches just using door closing events to develop the learning model).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Jun to incorporate the teachings of DiCarlo in order to detect vehicle door closing events and to “increase the accuracy of detecting door closing events while reducing the occurrence of false positive and false negative door closing detection” (par. 51). Learning how to identify a door slam would also include learning an expected acceleration range for a door slam. Detecting an abnormal value (for example, a non-door slam acceleration value) to determine a possible crash is well-known in the art (examples include Hyuk (KR 20110005994), Agata US 20220410675)), and would have been an obvious modification.
DiCarlo fails to teach and upon determining that the door changes from the open state to the closed state, learning the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point.
DiCarlo instead only teaches using “training sets of known gyroscope and/or sound level data associated with door closing events to the model” (par. 59). This training set data would presumably have been gathered from actual door closing data, which would have been gathered during a learning period between a first time point, at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point.
However, Hong teaches upon determining that the door changes from the open state to the closed state, learning the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point (par. 16, “the learning process of the machine learning model used to estimate that an impact caused by opening or closing a door or an impact other than door opening and closing occurs in the parked vehicle, i) the parking An impact test caused by opening or closing a vehicle door and an impact test other than door opening and closing are performed to collect acceleration data and air pressure data of the vehicle before and after the occurrence of each impact, and record the type of each impact”).
Hong teaches that these impact tests are done while a door is opened and closed. The machine learning model would need to be taught that an impact is done by a door closing event, therefore it would be determined whether or not a door is changing from an open to a closed state. Although it does not explicitly teach that the machine learning model is determining a threshold, in order to categorize the impact as either an impact caused by opening or closing the door or an impact caused by something else, the machine learning model would need to determine some sort of categorization threshold to differentiate between the two type of impacts. Since the model uses a characteristic element from the acceleration data, this threshold would need to be related to the acceleration data. It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Jun in view of DiCarlo to incorporate the teachings of Hong to include the learning process during an opening and closing of the door. Repeatedly learning can improve the machine learning model accuracy in determining a door closing event (par. 35).
Regarding claim 5, Jun teaches a non-transitory recording medium (Fig. 1, storage 120) having recorded thereon a computer program for causing a processor (processor 140) to execute a processing comprising:
determining whether an impact equal to or greater than an acceleration threshold value stored in a memory (par. 43, “The processor 140 may determine whether an impact of a predetermined reference value or more that is applied to the vehicle corresponds to an impact caused by an accident” is detected by an acceleration sensor (Fig. 1 and par. 50, “sensing device 200 may include an impact sensor”);
starting video recording (Fig. 3, S205 perform impact image recording) in the memory when the impact equal or greater than the acceleration threshold value stored in the memory is detected by the acceleration sensor (par. 43, "The processor 140 may determine whether an impact of a predetermined reference value or more that is applied to the vehicle corresponds to an impact caused by an accident") while an ignition power supply of a vehicle is turned off (Fig. 2 S100, vehicle enters parking recording mode);
determining whether a door of the vehicle changes from an open state to a closed state (Fig. 2 S300, monitor door communication signal);
wherein (abstract, “a processor that determines whether an impact of a reference value or more applied to the vehicle is caused by an accident”—the impact is used to determine if an accident has occurred, so obviously the reference value will need to be a threshold value that would indicate an accident),
Jun fails to teach upon determining that the door changes from the open state to the closed state, learning the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point, wherein learning the acceleration threshold value so that the acceleration threshold value is greater than the impact detected by the acceleration sensor when the door changes from the open state to the closed state, and less than or equal to the impact detected by the acceleration sensor when an accident occurs, and not using the impact detected by the acceleration sensor and associated with occurrence of the accident during the learning period in learning of the acceleration threshold value.
Jun instead only teaches checking whether the impact is a predetermined reference value or more, and does not teach how the reference value is determined. The reference value is also not used to differentiate between a door closing event and a collision event. Instead, it uses the door sensor to do so (Fig. 3, S300). However, using a threshold value to differentiate between a door closing event and a collision event is already well-known in the field.
DiCarlo teaches determining whether an impact equal to or greater than an acceleration threshold value stored in a memory (abstract, “determining an event signature based on the sensor data”; par. 46, “In some examples, the event signature may be characterized by the amplitude of initial impulse peak 514 exceeding a threshold value”),
(par. 59, “In some examples, a machine learning model may be applied to the event signature methods independently or in a combined manner. The machine learning model may be developed by providing training sets of known gyroscope and/or sound level data associated with door closing events to the model”),
wherein learning the acceleration threshold value so that the acceleration threshold value is greater than the impact detected by the acceleration sensor when the door changes from the open state to the closed state, and less than or equal to the impact detected by the acceleration sensor when an accident occurs (par. 59, “For example, if a vehicle receives an impact not due to a door closing (e.g. from an external object/person impacting the vehicle or from a passing truck creating an air pressure wave) an event signature from a gyroscope alone may falsely indicate a door closing event. A vehicle impact not due to a door closing event may generate a sound level signature distinguishable from a door closing sound level signature while the gyroscope data signature may not be able to distinguish between the door closing event and an impact due to another source. By correlating the gyroscope and sound level signatures a more robust (e.g. reducing false positive detection and/or false negative detection) event signature detection method is created”—DiCarlo teaches using both gyroscope data and sound data to differentiate between a door closing event and a collision event. Although not explicitly taught, the machine learning model would need to learn some kind of acceleration threshold that correlates with a sound level signature such that the acceleration teaches whether the event was a door closing event. It would be obvious to use a threshold that lies between a non-collision and a collision, and such thresholds are well-known in the field (examples include Hyuk (KR 20110005994), Agata US 20220410675)),
and not using the impact detected by the acceleration sensor and associated with occurrence of the accident during the learning period in learning of the acceleration threshold value (par. 59, “The machine learning model may be developed by providing training sets of known gyroscope and/or sound level data associated with door closing events to the model”—DiCarlo teaches just using door closing events to develop the learning model).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Jun to incorporate the teachings of DiCarlo in order to detect vehicle door closing events and to “increase the accuracy of detecting door closing events while reducing the occurrence of false positive and false negative door closing detection” (par. 51). Learning how to identify a door slam would also include learning an expected acceleration range for a door slam. Detecting an abnormal value (for example, a non-door slam acceleration value) to determine a possible crash is well-known in the art (examples include Hyuk (KR 20110005994), Agata US 20220410675)), and would have been an obvious modification.
DiCarlo fails to teach upon determining that the door changes from the open state to the closed state, learning the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point.
DiCarlo instead only teaches using “training sets of known gyroscope and/or sound level data associated with door closing events to the model” (par. 59). This training set data would presumably have been gathered from actual door closing data, which would have been gathered during a learning period between a first time point, at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point.
However, Hong teaches upon determining that the door changes from the open state to the closed state, learning the acceleration threshold value stored in the memory using the impact detected by the acceleration sensor during a learning period between a first time point at which the door changes from the open state to the closed state, and a second time point, which is a predetermined time before the first time point.
Hong teaches that these impact tests are done while a door is opened and closed. The machine learning model would need to be taught that an impact is done by a door closing event, therefore it would be determined whether or not a door is changing from an open to a closed state. Although it does not explicitly teach that the machine learning model is determining a threshold, in order to categorize the impact as either an impact caused by opening or closing the door or an impact caused by something else, the machine learning model would need to determine some sort of categorization threshold to differentiate between the two type of impacts. Since the model uses a characteristic element from the acceleration data, this threshold would need to be related to the acceleration data. It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Jun in view of DiCarlo to incorporate the teachings of Hong to include the learning process during an opening and closing of the door. Repeatedly learning can improve the machine learning model accuracy in determining a door closing event (par. 35).
Claim(s) 6-8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jun in view of DiCarlo and Hong, and further in view of Browne (US 20110270792 A1).
Regarding claim 6, the combination of Jun, DiCarlo, and Hong teaches the drive recorder device according to claim 1. Jun, DiCarlo, and Hong fail to teach the processor is further configured to perform the learning of the acceleration threshold value stored in the memory by using a largest acceleration or impact among accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and store the largest acceleration or impact in the memory as the acceleration threshold value after the learning.
Jun, DiCarlo, and Hong do not explicitly teach how a threshold would be determined via the machine learning model. However, using the largest data point would be an obvious and basic method of determining a threshold.
Browne teaches the processor is further configured to perform the learning of the acceleration threshold value stored in the memory by using a largest acceleration or impact among accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and store the largest acceleration or impact in the memory as the acceleration threshold value after the learning (par. 13, “Certain embodiments perform hypothesis testing by receiving a data stream and determining a noise subset of the data stream. The embodiments identify a maximum element from each of plural portions of the noise subset, select one of the identified maximum elements to generate a threshold, and obtain a decision stream based on the data stream compared to the threshold”).
Browne teaches a method to “discern, under conditions of uncertainty, whether or not a signal of interest exists in data by making decisions. In such problems a threshold test is often employed to partition data believed to contain the signal of interest from data believed to contain only noise” (par. 4), and teaches a way to minimize false positives or negatives (par. 3-8). Using the largest data point is a known method of determining a threshold and would have been an obvious option with reasonable expectation of success.
Regarding claim 7, the combination of Jun, DiCarlo, and Hong teaches the video recording method according to claim 4. Jun, DiCarlo, and Hong fail to teach performing the learning of the acceleration threshold value stored in the memory by using a largest acceleration or impact among accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and storing the largest acceleration or impact in the memory as the acceleration threshold value after the learning.
Jun, DiCarlo, and Hong do not explicitly teach how a threshold would be determined via the machine learning model. However, using the largest data point would be an obvious and basic method of determining a threshold.
Browne teaches performing the learning of the acceleration threshold value stored in the memory by using a largest acceleration or impact among accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and storing the largest acceleration or impact in the memory as the acceleration threshold value after the learning (par. 13, “Certain embodiments perform hypothesis testing by receiving a data stream and determining a noise subset of the data stream. The embodiments identify a maximum element from each of plural portions of the noise subset, select one of the identified maximum elements to generate a threshold, and obtain a decision stream based on the data stream compared to the threshold”).
Browne teaches a method to “discern, under conditions of uncertainty, whether or not a signal of interest exists in data by making decisions. In such problems a threshold test is often employed to partition data believed to contain the signal of interest from data believed to contain only noise” (par. 4), and teaches a way to minimize false positives or negatives (par. 3-8). Using the largest data point is a known method of determining a threshold and would have been an obvious option with reasonable expectation of success.
Regarding claim 8, the combination of Jun, DiCarlo, and Hong teaches the non-transitory recording medium according to claim 5. Jun, DiCarlo, and Hong fail to teach the computer program causes the processor to execute the processing further comprising, performing the learning of the acceleration threshold value stored in the memory by using a largest acceleration or impact among accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and storing the largest acceleration or impact in the memory as the acceleration threshold value after the learning.
Jun, DiCarlo, and Hong do not explicitly teach how a threshold would be determined via the machine learning model. However, using the largest data point would be an obvious and basic method of determining a threshold.
Browne teaches the computer program causes the processor to execute the processing further comprising, performing the learning of the acceleration threshold value stored in the memory by using a largest acceleration or impact among accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and storing the largest acceleration or impact in the memory as the acceleration threshold value after the learning (par. 13, “Certain embodiments perform hypothesis testing by receiving a data stream and determining a noise subset of the data stream. The embodiments identify a maximum element from each of plural portions of the noise subset, select one of the identified maximum elements to generate a threshold, and obtain a decision stream based on the data stream compared to the threshold”).
Browne teaches a method to “discern, under conditions of uncertainty, whether or not a signal of interest exists in data by making decisions. In such problems a threshold test is often employed to partition data believed to contain the signal of interest from data believed to contain only noise” (par. 4), and teaches a way to minimize false positives or negatives (par. 3-8). Using the largest data point is a known method of determining a threshold and would have been an obvious option with reasonable expectation of success.
Claim(s) 9-11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jun in view of DiCarlo and Hong, and further in view of Yang (Yang, J., Rahardja, S., & Fränti, P. (2019). Outlier detection. Proceedings of the International Conference on Artificial Intelligence, Information Processing and Cloud Computing - AIIPCC ’19. https://doi.org/10.1145/3371425.3371427).
Regarding claim 9, the combination of Jun, DiCarlo, and Hong teaches the drive recorder device according to claim 1. Jun, DiCarlo, and Hong fail to teach the processor is further configured to perform the learning of the acceleration threshold value stored in the memory by using a standard deviation of accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and store the standard deviation in the memory as the acceleration threshold value after the learning.
Jun, DiCarlo, and Hong do not explicitly teach how a threshold would be determined via the machine learning model. However, using the standard deviation would be an obvious and basic method of determining a threshold.
Yang teaches that using a standard deviation is one of the most common methods of determining a threshold (pg. 1, “We searched from Google Scholar using keyword “outlier detection” and it returned 38,900 related publications during the last 2 years (June 2016 to June 2018). We randomly picked 100 publications and studied what thresholding techniques were used. The results are summarized in Figure 1. They show that standard deviation, median absolute deviation and interquartile range (IQR) are the most used techniques”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Jun, DiCarlo, and Hong to have configured the processor to perform the learning of the acceleration threshold value stored in the memory by using a standard deviation of accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and store the standard deviation in the memory as the acceleration threshold value after the learning. The combination of Jun, DiCarlo, and Hong teaches determining whether an impact was caused by a door closing or a collision. A collision would be considered an outlier, and Jun, DiCarlo, and Hong teach detecting such an outlier. While it is not explicitly taught by Jun, DiCarlo, or Hong, Yang teaches that using the standard deviation to determine a threshold for outlier detection is a well-known method and would have been an obvious option with reasonable expectation of success.
Regarding claim 10, the combination of Jun, DiCarlo, and Hong teaches the video recording method according to claim 4. Jun, DiCarlo, and Hong fail to teach performing the learning of the acceleration threshold value stored in the memory by using a standard deviation of accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and storing the standard deviation in the memory as the acceleration threshold value after the learning.
Jun, DiCarlo, and Hong do not explicitly teach how a threshold would be determined via the machine learning model. However, using the standard deviation would be an obvious and basic method of determining a threshold.
Yang teaches that using a standard deviation is one of the most common methods of determining a threshold (pg. 1, “We searched from Google Scholar using keyword “outlier detection” and it returned 38,900 related publications during the last 2 years (June 2016 to June 2018). We randomly picked 100 publications and studied what thresholding techniques were used. The results are summarized in Figure 1. They show that standard deviation, median absolute deviation and interquartile range (IQR) are the most used techniques”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Jun, DiCarlo, and Hong to have configured the processor to perform the learning of the acceleration threshold value stored in the memory by using a standard deviation of accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and store the standard deviation in the memory as the acceleration threshold value after the learning. The combination of Jun, DiCarlo, and Hong teaches determining whether an impact was caused by a door closing or a collision. A collision would be considered an outlier, and Jun, DiCarlo, and Hong teach detecting such an outlier. While it is not explicitly taught by Jun, DiCarlo, or Hong, Yang teaches that using the standard deviation to determine a threshold for outlier detection is a well-known method and would have been an obvious option with reasonable expectation of success.
Regarding claim 11, the combination of Jun, DiCarlo, and Hong teaches the non-transitory recording medium according to claim 5. Jun, DiCarlo, and Hong fail to teach the computer program causes the processor to execute the processing further comprising, performing the learning of the acceleration threshold value stored in the memory by using a standard deviation of accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and storing the standard deviation in the memory as the acceleration threshold value after the learning.
Jun, DiCarlo, and Hong do not explicitly teach how a threshold would be determined via the machine learning model. However, using the standard deviation would be an obvious and basic method of determining a threshold.
Yang teaches that using a standard deviation is one of the most common methods of determining a threshold (pg. 1, “We searched from Google Scholar using keyword “outlier detection” and it returned 38,900 related publications during the last 2 years (June 2016 to June 2018). We randomly picked 100 publications and studied what thresholding techniques were used. The results are summarized in Figure 1. They show that standard deviation, median absolute deviation and interquartile range (IQR) are the most used techniques”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Jun, DiCarlo, and Hong to have configured the processor to perform the learning of the acceleration threshold value stored in the memory by using a standard deviation of accelerations or impacts detected multiple times by the acceleration sensor during the learning period from the second time point to the first time point, and store the standard deviation in the memory as the acceleration threshold value after the learning. The combination of Jun, DiCarlo, and Hong teaches determining whether an impact was caused by a door closing or a collision. A collision would be considered an outlier, and Jun, DiCarlo, and Hong teach detecting such an outlier. While it is not explicitly taught by Jun, DiCarlo, or Hong, Yang teaches that using the standard deviation to determine a threshold for outlier detection is a well-known method and would have been an obvious option with reasonable expectation of success.
Claim(s) 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jun in view of DiCarlo and Hong, and further in view of Mase (JP 2010228524 A).
Regarding claim 15, the combination of Jun, DiCarlo, and Hong teaches the drive recorder device according to claim 1. Jun, DiCarlo, and Hong fail to teach the processor is configured to determine whether the door of the vehicle changes from the open state to the closed state in response to the acceleration sensor detecting an impact smaller than the acceleration threshold value stored in the memory.
However, Mase teaches wherein the processor is configured to determine whether the door of the vehicle changes from the open state to the closed state in response to the acceleration sensor detecting an impact smaller than the acceleration threshold value stored in the memory (par. 43 and Fig. 4, “Note that the side collision determination unit 104 repeats the processing from step S401 onward when it is determined that the door portion is in the open state (NO in step S401) or when acceleration is not detected (NO in step S402)”; par. 48 and Fig. 4, “On the other hand, when it is determined that the detected acceleration value is equal to or less than the predetermined value (NO in step S406), the side collision determination unit 104 repeats the processing in step S401 and subsequent steps”).
Mase teaches a method for discriminating between a side collision and door closing which includes continuously checking whether or not the door is open. Mase is specifically directed towards preventing an airbag from deploying due to a door closing. It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Jun, DiCarlo, and Hong to incorporate the teachings of Mase to teach the method continuously loops even when an impact does not meet the threshold and to continuously check if the door is open or not. Mase teaches that by determining if a door is open, it is able to better prevent from erroneous discrimination between a door closing event and a collision (par. 8).
Allowable Subject Matter
Claims 12-14 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
The use of DBSCAN, while a known method for determining outliers, is not necessarily obvious for the purpose of removing an occurrence of an accident as noise during learning of the acceleration threshold value.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Hyuk (KR 20110005994) and Agata US 20220410675) teach using a threshold to determine outliers in door slamming events or collision events.
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MINATO LEE HORNER whose telephone number is (571)272-5425. The examiner can normally be reached M-F 8-5.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Christian Chace can be reached at (571) 272-4190. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/M.L.H./Examiner, Art Unit 3665 /CHRISTIAN CHACE/Supervisory Patent Examiner, Art Unit 3665
Prosecution Insights