Multiple exposure event determination
Systems, devices and methods provide, implement, and use vision-based methods of sequence inference for a device affixed to a vehicle.
1. A computer-implemented method comprising:
receiving, by a computer, video data from a camera on a vehicle;
detecting, by the computer, at least two objects in a plurality of frames of the video data, the plurality of frames corresponding to a duration of a detected traffic event;
converting, by the computer, an encoding of the detected at least two objects into a fixed-length input, the encoding having a length corresponding to a number of the at least two objects detected in each frame of the plurality of frames;
determining, by the computer, using the fixed-length input, an inference of the vehicle's actions in the traffic event; and
classifying, by the computer, a driver behavior as safe based on the inference.
2. The method according to claim 1 , wherein the computer applies a deep neural network trained to determine the inference based on an input having a length of the fixed-length input.
3. The method according to claim 2 , wherein the deep neural network is trained using supervised training.
4. The method according to claim 2 , further comprising accessing, by the computer for the classification, at least one of: GPS location during the duration, speed measurements during the duration, heading measurements during the duration, inertial measurements during the duration, or lane detections during the duration.
5. The method according to claim 1 , wherein the duration corresponds to a first instance of a first object of the at least two objects in a first frame extending a sequence of frames to a second instance of the first object of the at least two objects in a second frame.
6. The method according to claim 1 , wherein detecting the at least two objects comprises applying the video data to a neural network configured to output a bounding box for each of the at least two objects.
7. The method according to claim 6 , wherein a confidence of detecting a first object of the at least two objects in a frame is increased when the first object of the at least two objects is detected in a preceding frame and a following frame.
8. The method according to claim 6 , wherein a confidence of detecting a first object of the at least two objects in a frame is decreased when the first object of the at least two objects is not detected in a preceding frame and a following frame.
9. The method according to claim 1 , further comprising generating, by the computer, a feature vector for each frame during the duration.
10. The method according to claim 1 , wherein the fixed-length input comprises a single input data vector of a predetermined size.
11. The method according to claim 10 , wherein the single input data vector contains a superposition of inferences from the plurality of frames.
12. A system comprising:
a non-transitory computer-readable medium storing instructions; and
at least one processor configured to execute the instructions to:
receive video data from a camera on a vehicle;
detect at least two objects in a plurality of frames of the video data, the plurality of frames corresponding to a duration of a detected traffic event;
convert an encoding of the detected at least two objects into a fixed-length input, the encoding having a length corresponding to a number of the at least two objects detected in each frame of the plurality of frames;
determine, using the fixed-length input, an inference of the vehicle's actions in the traffic event; and
classify a driver behavior as safe based on the inference.
13. The system according to claim 12 , wherein the at least one processor is further configured to apply a deep neural network trained to determine the inference based on an input having a length of the fixed-length input.
14. The system according to claim 13 , wherein the deep neural network is trained using supervised training.
15. The system according to claim 12 , wherein the duration corresponds to a first instance of a first object the at least two objects in a first frame extending a sequence of frames to a second instance of the first object of the at least two objects in a second frame.
16. The system according to claim 12 , wherein a confidence of detecting a first object of the at least two objects in a frame is increased when the first object of the at least two objects is detected in a preceding frame and a following frame.
17. The system according to claim 12 , wherein a confidence of detecting a first object of the at least two objects in a frame is decreased when the first object of the at least two objects is not detected in a preceding frame and a following frame.
18. The system according to claim 12 , wherein the at least one processor is further configured to generate a feature vector for each frame during the duration.
19. The system according to claim 12 , wherein the fixed-length input comprises a single input data vector of a predetermined size.
20. The system according to claim 19 , wherein the single input data vector contains a superposition of inferences from the plurality of frames.