Performance of video detectors working with lossy compressed video streams

Complex traffic control systems are equipped with a range of cameras for traffic surveillance, road traffic measurements. On many sites the different cameras cover the same observation areas but provide different quality streams to the system, usually compressed for surveillance and raw for vehicle detection. Elimination of duplicate cameras especially high quality devices is desired for enhancing the performance of systems. Vehicle detectors based on image processing are sensitive to the quality of input video streams. The paper presents results from tests of using lossy data compression for delivering video streams to vehicle detectors for traffic control. The limit of data loss is determined for assuring correct vehicle detection. The recommendations can be used for optimising traffic vision systems.