Visual tracking based on stacked Denoising Autoencoder network with genetic algorithm optimization

Visual object tracking in dynamic environments with severe appearance variations is a significant problem in the computer vision field. This paper proposes a novel visual tracking algorithm that exploits the multiple level features learning ability of SDAE. There are two training stages for the SDAE network: Layer-wise pre-training and fine-tuning. In the pre-training stage, a two-layer sparse-coded method is used to represent the input image, then a multi-level image feature descriptor is obtained. In the fine-tuning stage, the connection weights and bias terms for back propagation are gathered via genetic algorithm. A logistic classification layer is added at the top of the encoder network to enable tracking within the well-established particle filter network. Experimental results confirm, both qualitatively and quantitatively, that the proposed method performs well in comparison against eight other state-of-the-art methods.