Revision history [back]

If you are working with video, you can track the vehicle bounding boxes from frame to frame.

In each frame, track the location and velocity of the the vehicle.

Vehicle location = a(prediction from previous frame) + (1-a)current_frame_location prediction = current_location + t*velocity velocity = (current_location - previous_location)/t.

You should add code for creating new tracks - when you have a new detection that doesn't fit any of the current tracks. You should also add code for deleting "dead tracks" - when the car is no longer in the current frame.

Please tell me if you have any questions, and I'd try to make my answer clearer.

If you are working with video, you can track the vehicle bounding boxes from frame to frame.

In each frame, track the location and velocity of the the vehicle.

Vehicle location = a(prediction from previous frame) + (1-a)current_frame_location current_frame_location

prediction = current_location + t*velocity t*velocity

velocity = (current_location - previous_location)/t.

You should add code for creating new tracks - when you have a new detection that doesn't fit any of the current tracks. You should also add code for deleting "dead tracks" - when the car is no longer in the current frame.

Please tell me if you have any questions, and I'd try to make my answer clearer.