Extracting traffic flow velocity and vehicle density within MPEG videos using motion vectors and background estimation technique

Metropolitan Municipality of Istanbul establishes a foundation called ?Traffic Control Centre?. This foundation has nearly 150 traffic cameras (in 2006) at the important points of Istanbul for watching the traffic situation. Operators watch this cameras scene continuously, they declare situation if undesirable event occurs. This system is occur waste of manpower, and cause of depending manpower, there may be a lot of errors. During thesis, developed system extract two most important parameter from traffic cameras are vehicle density and average vehicle velocity. By this mean we provide data for traffic planning automatically. Average vehicle velocity is extracted by motion vector which exist in MPEG format. By this mean optical flow is found by less complexity algorithm. Geometric correction was implemented to this motion vector for finding real motion. Background estimation which is the most important step for finding average vehicle density was calculated by probability based model that called Gaussian Mixture Model. For calculate foreground, basic subtraction to current image from background image and some noisy filter was implemented. Ratio of foreground and background was declared average vehicle density. For measuring the performance of developed system, used nearly 200 minutes traffic cameras record which is belong 10 different cameras. Velocity extraction performs every environmental situation under %10 error rate, but average vehicle density extraction is not successful at night, or variable light conditions as cloudy days. Keywords: Traffic Flow, Velocity Measurement, MPEG, Traffic Parameters, Vehicle Density, Geometric Correction, Background Estimation