An ongoing effort to automate the tracking of vehicles in video taped images is described. The first stages of such a system have been implemented. The concept of using the two dimensional projection of a vehicle onto the road surface to create a "footprint" silhouette is developed. The feasibility and the software requirements to use the footprinting method to measure vehicle positions in video scenes is described. Positions of vehicles found with the footprinting method were compared to those found by the conventional method of marking the tire positions of vehicles. There were considerable differences between positions reported using the two different methods. The differences were mainly due to the accuracy of the calibration coefficients used to define transform equations which transform positions from the image plane onto the road surface. the footprinting method is shown to work, however numerous suggestions are made for improvement of the technique. The development of the system consists of the detection of moving vehicles using the difference between two successive video frames. Seven different software routines have been completed and applied to some traffic scenes to demonstrate the differencing techniques. Last, a list of tasks are presented, which when completed, will make up a software system that is able tot automatically track vehicles in video scenes. (Author/publisher)
Abstract