Correlation filter based fisheye video target tracking with adaptive weighted feature integration

Tracking target in fisheye camera is getting increasing interest and wide applications. However, due to the special imaging principle of the fisheye lens, videos shoot by fisheye cameras have serious distortion, which brings great interference to the target tracking. In recent years, correlation filters have gained wide attention with its fast and robust characteristics, and have gradually become a type of important method in the field of target tracking. The proposal of the kernelized correlation filter makes a large number of target features available for tracking enhancements. In order to meet the real-time, this paper creatively employs correlation filters for target tracking on fisheye videos. Aiming at the distortion handling, this paper proposes a feature integration with adaptive weight updating and incorporates this feature into the kernelized correlation filtering method. Moreover, this paper makeups a fisheye video data set for evaluating tracking performance. The evaluation results validate that the proposed approach can greatly reduce the impact of deformation on tracking on the basis of the real-time, and also obtain good tracking performance.