Statistical approach for detection of vehicle in heavy traffic

In this thesis an innovative system for detecting and extracting vehicles in traffic surveillance scenes is presented .The main concept behind vehicle detection in a live video is extract the foreground and remove the background from it .This theory is called background subtraction .This method can be implemented in various ways such as setting a particular threshold value and removing the objects having value less than it .The second approach is to compare the current frame with the previous frame and if the variance is more than a certain value it detects the motion of that object .Third and the most efficient method is to use a statistical method where a certain number of video frames are used to initialize a fixed number of Gaussian modes in the mixture model. While in the first method only white cars are being detected this disadvantage is solved when we use a statistical method where a particular vehicle is detected using a foreground detection technique on a frame .Here the input video file is read in AVI format .After that morphological operations are done on it and the bounding box is calculated .Finally the moving object is presented with a rectangle drawn around it and total number of vehicles in the current frame is calculated .This process is repeated for each frame till the whole video is processed .Since this method uses a training set and not a general threshold selected manually by the user the foreground extracted is more desirable than other method and besides it requires much less memory than the method where the background subtraction is done by comparing the frame with the previous one .And last but not the least it gives a general idea about the vehicle frequency in the video which can be very helpful in traffic monitoring