Surveillance from above: A detection-and-prediction based multiple target tracking method on aerial videos

Along with the rapid development of the Unmanned Aerial Vehicle (UAV) technology, visual traffic surveillance is now granted with a great opportunity to become mobilized. Airborne mounted surveillance cameras are suitable to work as an ideal alternative to the traditional stationary cameras to provide more coverage with lower expense. For realizing such a solution, in this paper we engineered an efficient multiple targets tracking method designed for aerial videos captured by low-altitude UAVs. This method combines the strengths of both Tracking-by-Detection (T-by-D) and Tracking-by-Prediction (T-by-P) framework, where the target detection is first performed with an integrated motion detection module, then the traditional data association component is substituted by a detection-and-prediction based category-free blob tracking scheme. To overcome the camera jitter introduced by the UAV platform, a reliable image registration method is also implemented as a pre-processing operation. Extensive experiments have demonstrated the effectiveness and reliability of the proposed method.