ROBUST OBJECT TRACKING AND ADAPTIVE DETECTION FOR AUTO NAVIGATION OF UNMANNED AERIAL VEHICLE

Abstract

Abstract Object detection and tracking is an important research topic in the computer vision field with numerous practical applications. Although great progress has been made, both in object detection and tracking over the past decade, it is still a big challenge in real-time applications like automated navigation of an unmanned aerial vehicle and collision avoidance with a forward looking camera. An automated and robust object tracking approach is proposed by integrating a kernelized correlation filter framework with an adaptive object detection technique based on minimum barrier distance transform. The proposed tracker is automatically initialized with salient object detection and the detected object is localized in the image frame with a rectangular bounding box. An adaptive object redetection strategy is proposed to refine the location and boundary of the object, when the tracking correlation response drops below a certain threshold. In addition, reliable pre-processing and post-processing methods are applied on the image frames to accurately localize the object. Extensive quantitative and qualitative experimentation on challenging datasets have been performed to verify the proposed approach. Furthermore, the proposed approach is comprehensively examined with six other recent state-of-the-art¬ trackers, demonstrating that the proposed approach greatly outperforms these trackers, both in terms of tracking speed and accuracy.