Multi-object tracking method based on super-pixel and energy minimization

Video target tracking research is a hot topic in the field of computer vision. It has broad prospects in many fields such as video surveillance, virtual reality, interactive and automatic navigation. For multi-object tracking in video sequences, we propose a method based on super pixel and energy minimization. First we achieve target detection by super-pixel labeling with minimized energy function, and we capture the target's appearance with data cost model, which can be trained by support vector machine (SVM). Next we can track object based on continuous energy minimization method. The recall ratio of this method can be 99.8% at high overlap rate, e.g., 0.9, and yields improvement by 70%. Experimental results show that the method has advantages over other methods in complex environment with similar color background and targets, and improves the accuracy of multiple target tracking.