A Scale Adaptive and Anti-occlusion Tracking Algorithm with Feature Fusion

KCF is an excellent target tracking algorithm with fast computing speed and high accuracy. However, it performs poorly in complex tracking situations such as target deformation, motion blur, scale change and occlusion. In view of target deformation and motion blur, we designed a feature fusion method, which combines CN feature and HOG feature to enhance the expression ability of the model. In view of the change of scale, the scale pool is designed. To improve the ability of anti occlusion, we improved the model updating mechanism and designed a SVM detector to detect the target after it is lost. The experiments on OTB-100 showed that the improved method achieves a great improvement compared with KCF, the accuracy increases by 4.2%, the success rate increases by 12.1%, and our algorithm meets the real-time requirements.