Handcrafted and Deep Trackers: Recent Visual Object Tracking Approaches\n and Trends

摘要

In recent years visual object tracking has become a very active research\narea. An increasing number of tracking algorithms are being proposed each year.\nIt is because tracking has wide applications in various real world problems\nsuch as human-computer interaction, autonomous vehicles, robotics, surveillance\nand security just to name a few. In the current study, we review latest trends\nand advances in the tracking area and evaluate the robustness of different\ntrackers based on the feature extraction methods. The first part of this work\ncomprises a comprehensive survey of the recently proposed trackers. We broadly\ncategorize trackers into Correlation Filter based Trackers (CFTs) and Non-CFTs.\nEach category is further classified into various types based on the\narchitecture and the tracking mechanism. In the second part, we experimentally\nevaluated 24 recent trackers for robustness, and compared handcrafted and deep\nfeature based trackers. We observe that trackers using deep features performed\nbetter, though in some cases a fusion of both increased performance\nsignificantly. In order to overcome the drawbacks of the existing benchmarks, a\nnew benchmark Object Tracking and Temple Color (OTTC) has also been proposed\nand used in the evaluation of different algorithms. We analyze the performance\nof trackers over eleven different challenges in OTTC, and three other\nbenchmarks. Our study concludes that Discriminative Correlation Filter (DCF)\nbased trackers perform better than the others. Our study also reveals that\ninclusion of different types of regularizations over DCF often results in\nboosted tracking performance. Finally, we sum up our study by pointing out some\ninsights and indicating future trends in visual object tracking field.\n