Vehicle classification and lane categorization

This research was carried out in order to use image processing techniques to detect, track, classify, and count vehicles in a real-time traffic video. Additionally this research also focuses on occlusion detection and handling, which involves detecting and handling the states when one vehicle superimposes the other or when two superimposed vehicles split, and lane categorization i.e. dividing the vehicles in the moving traffic scene into their corresponding lanes of the road in which they are travelling in. The challenge taken up was to achieve all this using basic image processing techniques and algorithms and not compromise on the desired output, most importantly keeping the single frame processing time less and the speed of processed video playback to be comparable to that real-time video playback speeds. It is to note that all of the above is achieved through the techniques demonstrated in this research. The process cycle followed from frame extraction to the final classification of vehicles involves algorithms that are all highly region based and do the necessary processing only on the region of interest, only as and when required. Thus saving processing power and time. The final results obtained could easily be molded to use in any application in surveillance, traffic monitoring etc. by just adding the required decision-making statements in it.