Target positioning of multiple cameras for dense crowd based on line-of-sight constraint

For the target positioning under multi-camera setup, the key problem is how to realize targets correspondence between cameras. Two main popular methods are epipolar geometric constraint and homography matrix constraint. However, both of them have a low performance in the multi-camera surveillance system because of large baseline and large angle of view. In the paper, we proposed a correspondence algorithm based-on the constraint of line-of-sight for the crowd positioning. Based on the detection of head region, accuracy correspondence could be achieved by using the line-of-sight distance. The experimental results based on experimental moving data show that the method has a high accuracy and a lower computational complexity.