Handcrafted and Deep Trackers: A Survey

Visual object tracking is a very interesting research field in computer vision and artificial intelligence with a growing number of proposed trackers each year. Later, owing to the advancement of deep learning methodologies, trackers based on deep networks have obtained a significant interest from many researchers due to their excellent tracking performance. The main objectives of this paper include, firstly, we give a comprehensive review of the recent proposed trackers. According to the feature descriptors used to represent the target and tracking architecture, we group the selected tracking algorithms into two main categories: handcrafted trackers and deep trackers, each of which has several subcategories. Secondly, we evaluate 22 recent trackers for their respective strengths and weaknesses then we compare their tracking performance. At last, we conclude our review by outlining the main obtained results, future directions and suggestions for developing newly proposed trackers. Our study shows that trackers based on both handcrafted and deep features achieved best tracking results.