Multi-Object Tracking Based on Deep Probability Fusion with Holistic Feature Reprentation and IMM-JIPDA

Among the difficult problem of multi-object surface feature representation, in order to better realize the feature representation method of multi-object and enhance the discrimination of multi-object, a new model of multi-object tracking surface feature representation based on deep probability fusion with holistic feature representation and Interacting multiple model with interacting multiple model joint integrated probabilistic data association is proposed, which is more consistent with the overall video scene understanding. We test the proposed method on data set Multiple Object Tracking 16. The experiment shows that our method has improvements in performance.