New region-based surveillance scheme for high-way multi-vehicle recognition and tracking
Qiang Liu · Guangdian gongcheng · 2006
This paper provides a new region-based high-way multi-vehicle surveillance system scheme. The contribution of this system includes a modified adaptive Gaussian background model, a non-recursive region-growth algorithm for multi-vehicle recognition, and a color-based predictive rule for multi-vehicle tracking. Since the high-way images are usually strongly disturbed, a modified region-based Gaussian background model is presented in order to remove much heavy noise and small disturbing regions. To build this model, we first used several background imaging samples to get the disturbed regions by background subtraction technique, and then computed the statistic value of the size of disturbed regions as the parameters of this region-based Gaussian background model. In order to reduce the effect of daylight and be updated adaptively, we also applied a weighted iterative algorithm based on Kalman filter theory. A modified non-recursive region-growth algorithm is presented for multi-vehicle recognition whose computational complexity can be only O(n), and a color-based predictive rule is also proposed for multi-vehicle tracking which is very easy to implement. At last, the experimental results show that this system can achieve multi-vehicle recognition and tracking robustly and adaptively.