Reducing Frame Rate for Object Tracking

Abstract

AbstractObject tracking is commonly used in video surveillance, but typically video with full frame rate is sent. We previously have shown that full frame rate is not needed, but it is unclear what the appropriate frame rate to send or whether we can further reduce the frame rate. This paper answers these questions for two commonly used object tracking algorithms (frame-differencing-based blob tracking and CAMSHIFT tracking). The paper provides (i) an analytical framework to determine the critical frame rate to send a video for these algorithms without them losing the tracked object, given additional knowledge about the object and key design elements of the algorithms, and (ii) answers the questions of how we can modify the object tracking to further reduce the critical frame rate. Our results show that we can reduce the 30 fps rate by up to 7 times for blob tracking in the scenario of a single car moving across the camera view, and by up to 13 times for CAMSHIFT tracking in the scenario of a face moving in different directions.KeywordsAverage ErrorFrame RateTracking AlgorithmObject TrackingCurrent FrameThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.