Multi-object tracking based on thermal-visible video sequence fusion

In this work, we firstly propose a two direction two dimensional principal component analysis based fusion framework that addresses the problems of thermal-visible video sequence fusion. Then, the thermal filter is designed for target detection based on the characteristics of the color fusion image. The regions of interest that include both of thermal and visible information of motion object. Finally, multiple object tracking and classification implement uses locality preserving projection to maintain the structure of the manifold and reduce the dimension of the target module. Results demonstrate the advantage of our proposed method in obtaining accurately results for both image fusion and tracking.