Target Tracking by a Single Camera Based on Range-Window Algorithm and Pattern Matching: Real Time Operation 2007-01-1320
A method, which determines the range and lateral position of a preceding vehicle on the road by a single image, had been proposed. It is based on the Range-Window algorithm (RWA) and Pattern Matching (PM). The RWA estimates the range by using multiple virtual windows of fixed physical size at different distances. The size ratio between the windows and a preceding vehicle determines the best window. The associated range of the window will be the range of the vehicle. The PM complementarily estimates the range by using a template obtained through the RWA. It works especially well when an occlusion occurs due to shadows of road side objects. The range estimation is based on the use of horizontally-modified patterns of the template. Namely this PM can perform the range estimation as well as the object extraction. Here, for the real time operation, the calculation time of this method was evaluated after coding the RWA and PM into C-language (16 bit calculation) from Matlab (64 bit). In an off-line test using 5,500 images, the time for an image of 320 × 240 cells is less than 0.1 [sec], the detection rate was 96%, the range accuracy was better than 1.0 [m] and the lateral position accuracy was high. This was determined to be fast enough for typical radar applications and the performance degradation by bit reduction was not observed. This method exhibits a fast calculation time as well as acceptable performance. This method is expected to be applied to a low speed follower.