A robust object tracking synthetic structure using regional mutual information and edge correlation-based tracking algorithm in aerial surveillance application

This paper issues the problem of moving object tracking in aerial video sequences for surveillance application. The proposed object tracking synthetic structure integrates edge correlation-based (EC) tracking algorithm and a novel regional mutual information-based (RMI) tracking algorithm. In this structure, using a novel defined crowded criterion, we are able to recognize a crowded background and vast difference of illumination. The proposed crowded criterion determines the crowded and non-crowded backgrounds based on two defined intensity variation and relative power measures that are calculated from four rectangular areas around the object. Due to inability of the EC tracking algorithm in tracking the object in the crowded frames, the RMI tracking algorithm is applied to track an object in the crowded frames of video sequences. Against, due to sensitivity of the RMI tracking algorithm to scale variations of the object and less computational complexity of the EC tracking algorithm, the EC tracking algorithm is selected for the object tracking in the non-crowded frames of the video sequences. However, the RMI tracking algorithm is suitable for the frames with the crowded backgrounds, high-intensity variations, noisy conditions, and existence of clutter, but it is sensitive to scale variations of the object. Moreover, it is normally slower than the EC tracking algorithm, but we apply Powell-Golden optimization method to optimize the RMI tracking algorithm, until we can use it as an online algorithm on each frame. The obtained results show the superiority of the proposed tracking structure in comparison with other conventional algorithms. Our proposed structure covers the most challenges of the tracking in the aerial surveillance application.