Online object tracking based on sparse subspace representation
Baoyun Wang, Fei Chen, Ping Deng · 2014
In this paper, we propose an online object tracking algorithm, which combines incremental subspace learning with sparse representation. In the particle filter framework, we take Gaussian random sampling and use sub-sampling to filter the samples. We update the state of the training set through incremental PCA algorithm, then construct sparse subspace model using the eigenvectors of the training set. Before adding the tracking result into the training set, we adopt occlusion detection method to estimate. This paper implements a real-time tracking algorithm in various complex environments like deformation, rotation, illumination change and occlusion. Meanwhile, the tracking box can adjust with the scale and rotation of the object.