Abstract: To solve the problem of low vehicle tracking accuracy of traditional algorithms in
complex background, an improved method is proposed in this paper. It exploits vehicle tracking
polygon window and more exact prediction of search area to realize multiple feature matching
vehicle tracking, which is applied to detect vehicles retrograde motion. The experimental results
show that the algorithm has higher vehicle tracking accuracy while satisfying the need of real-time
and stability requirement than traditional ones.

Key words: vehicle tracking, multiple feature matching, complex background