Video analysis for efficient sorting of event data
A system for sorting event data includes an interface and a processor. The interface is configured to receive a small set of video frame data from a vehicle event recorder. The processor is configured to determine that the small set of video frame data indicates to fetch more vehicle event recorder data; determine a categorization based at least in part on the more vehicle event recorder data; and in response to the categorization, route the more vehicle event recorder data to a reviewing queue.
1. A system for sorting event data, comprising:
an interface configured to:
receive a small set of video frame data from a vehicle event recorder, wherein the small set of video frame data comprises one of a predetermined number of video frames and a predetermined seconds of video, wherein the small set of video frame data is determined based on an event being a first event or a second event;
a processor; and
a memory coupled with the processor, wherein the memory is configured to provide the processor with instructions which when executed cause the processor to:
determine, using a machine learning model, whether the small set of video frame data indicates that something unusual or important has occurred, wherein the machine learning model is trained using a portion of a set of examples, wherein the machine learning model is verified using a remaining portion of the set of examples, and wherein the machine learning model is tested using another portion of the set of examples, the other portion being different from the portion of the set of examples, the other portion being different from the remaining portion of the set of examples, wherein the determining of whether the small set of video frame data indicates that something unusual or important has occurred comprises to:
determine whether the small set of video frame data indicates that a vehicle pulls to a side of the road and then continues; and
in the event that the small set of video frame data indicates that a vehicle pulls to a side of the road and then continues:
determine whether the small set of video frame data indicates that a passenger was riding in the vehicle; and
in response to a determination that the small set of video frame data indicates that a passenger was riding in the vehicle, determine that the small set of video frame data indicates that something unusual or important has occurred; and
in response to a determination that the small set of video frame data indicates that something unusual or important has occurred:
determine that the small set of video frame data indicates to fetch more vehicle event recorder data from the vehicle event recorder;
receive the more vehicle event recorder data, wherein the more vehicle event recorder data comprises one or more of the following: more video frames, video data, GPS data, accelerometer data, brakes data, vehicle status data, lane departure data, collision avoidance data, and/or vehicle bus data;
determine, using machine vision analysis, whether the more vehicle event recorder data does not indicate that the event has occurred; and
in response to a determination that the more vehicle event recorder data does not indicate that an event has occurred, perform a review of the more vehicle event recorder data, wherein the event comprises at least one of the following: a lane departure, a lane drifting, and/or a rolling stop, and wherein the performing of the review of the more vehicle event recorder data comprises to:
determine a review type categorization based on the more vehicle event recorder data, wherein the review type categorization comprises one or more of: not interesting, human review, machine vision review, and/or collision review;
upon determining the review type categorization, route the more vehicle event recorder data to a reviewing queue; and
receive a review from the reviewing queue.
2. The system of claim 1 , wherein the processor is further to provide an indication to have the vehicle event recorder send the more vehicle event recorder data.
3. The system of claim 1 , wherein the small set of video frame data comprises N video frames.
4. The system of claim 3 , wherein N is 6.
5. The system of claim 1 , wherein to fetch more vehicle event recorder data, the processor is further to fetch S seconds of video.
6. The system of claim 5 , wherein S is 12.
7. The system of claim 1 , wherein the review type categorization comprises one of human review, machine vision review, movement analysis, or collision review.
8. The system of claim 1 , wherein the processor is further to determine whether to perform a review of the small set of video frame data.
9. The system of claim 8 , wherein in response to a determination to perform the review of the small set of video frame data, the processor is further to indicate to perform a review of the small set of video frame data.
10. The system of claim 1 , wherein the processor is further to determine whether an analysis of the more vehicle event recorder data indicates nothing unusual.
11. The system of claim 1 , wherein the processor is further to determine whether an analysis of the more vehicle event recorder data indicates an event.
12. A method for sorting event data, comprising:
receiving a small set of video frame data from a vehicle event recorder, wherein the small set of video frame data comprises one of a predetermined number of video frames and a predetermined seconds of video, wherein the small set of video frame data is determined based on an event being a first event or a second event;
determining, using a machine learning model, whether the small set of video frame data indicates that something unusual or important has occurred, wherein the machine learning model is trained using a portion of a set of examples, wherein the machine learning model is verified using a remaining portion of the set of examples, and wherein the machine learning model is tested using another portion of the set of examples, the other portion being different from the portion of the set of examples, the other portion being different from the remaining portion of the set of examples, wherein the determining of whether the small set of video frame data indicates that something unusual or important has occurred comprises:
determining whether the small set of video frame data indicates that a vehicle pulls to a side of the road and then continues; and
in the event that the small set of video frame data indicates that a vehicle pulls to a side of the road and then continues:
determining whether the small set of video frame data indicates that a passenger was riding in the vehicle; and
in response to a determination that the small set of video frame data indicates that a passenger was riding in the vehicle, determining that the small set of video frame data indicates that something unusual or important has occurred; and
in response to a determination that the small set of video frame data indicates that something unusual or important has occurred:
determining that the small set of video frame data indicates to fetch more vehicle event recorder data from the vehicle event recorder;
receiving the more vehicle event recorder data, wherein the more vehicle event recorder data comprises one or more of the following: more video frames, video data, GPS data, accelerometer data, brakes data, vehicle status data, lane departure data, collision avoidance data, and/or vehicle bus data;
determining, using machine vision analysis, whether the more vehicle event recorder data does not indicate that the event has occurred; and
in response to a determination that the more vehicle event recorder data does not indicate that an event has occurred, performing a review of the more vehicle event recorder data, wherein the event comprises at least one of the following: a lane departure, a lane drifting, and/or a rolling stop, and wherein the performing of the review of the more vehicle event recorder data comprises:
determining a review type categorization based on the more vehicle event recorder data, wherein the review type categorization comprises one or more of: not interesting, human review, machine vision review, and/or collision review;
upon determining the review type categorization, routing the more vehicle event recorder data to a reviewing queue; and
receiving a review from the reviewing queue.
13. A computer program product for sorting event data, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:
receiving a small set of video frame data from a vehicle event recorder, wherein the small set of video frame data comprises one of a predetermined number of video frames and a predetermined seconds of video, wherein the small set of video frame data is determined based on an event being a first event or a second event;
determining, using a machine learning model, whether the small set of video frame data indicates that something unusual or important has occurred, wherein the machine learning model is trained using a portion of a set of examples, wherein the machine learning model is verified using a remaining portion of the set of examples, and wherein the machine learning model is tested using another portion of the set of examples, the other portion being different from the portion of the set of examples, the other portion being different from the remaining portion of the set of examples, wherein the determining of whether the small set of video frame data indicates that something unusual or important has occurred comprises:
determining whether the small set of video frame data indicates that a vehicle pulls to a side of the road and then continues; and
in the event that the small set of video frame data indicates that a vehicle pulls to a side of the road and then continues:
determining whether the small set of video frame data indicates that a passenger was riding in the vehicle; and
in response to a determination that the small set of video frame data indicates that a passenger was riding in the vehicle, determining that the small set of video frame data indicates that something unusual or important has occurred; and
in response to a determination that the small set of video frame data indicates that something unusual or important has occurred:
determining that the small set of video frame data indicates to fetch more vehicle event recorder data from the vehicle event recorder;
receiving the more vehicle event recorder data, wherein the more vehicle event recorder data comprises one or more of the following: more video frames, video data, GPS data, accelerometer data, brakes data, vehicle status data, lane departure data, collision avoidance data, and/or vehicle bus data;
determining, using machine vision analysis, whether the more vehicle event recorder data does not indicate that the event has occurred; and
in response to a determination that the more vehicle event recorder data does not indicate that an event has occurred, performing a review of the more vehicle event recorder data, wherein the event comprises at least one of the following: a lane departure, a lane drifting, and/or a rolling stop, and wherein the performing of the review of the more vehicle event recorder data comprises: determining a review type categorization based on the more vehicle event recorder data, wherein the review type categorization comprises one or more of: not interesting, human review, machine vision review, and/or collision review;
upon determining the review type categorization, routing the more vehicle event recorder data to a reviewing queue; and
receiving a review from the reviewing queue.