Object tracking based on response maps fusion Siamese network

The full convolution Siamese network for object tracker formulate tracking as convolutional feature cross-correlation between a target template and a search region. This tracker realizes real-time object tracking. However, when there are interference factors similar to the target object, Siamese trackers still have an accuracy gap compared with state-of-theart algorithms. Therefore, we proposes an object tracking based on response maps fusion Siamese network(Siam-RMF ). Different from the full convolution Siamese network for object tracker, when the Siam-RMF tracker performs similarity learning, it no longer uses the features extracted by the last layer of the network, but extracts the features of the last three-layer network. Moreover, we propose a new model architecture to perform layer-wise and depth-wise aggregations, the depth-wise separable convolution is used to learn the similarity respectively to obtain the effective fusion of the corresponding depth cross-correlation response map. The fusion response maps can effectively avoid the loss of spatial information after multi-layer feature extraction. Experimental results on TB50 and UAV123L demonstrate the effectiveness of the proposed tracker without decreasing the tracking speed, and show stronger robustness and better tracking performance in complex environments.