NCTU

An UAV Autopilot System for Sports Player Tracking

2017/08/02
CodeM17-KWC-25-1
Professorprofessor Kuan-Wen Chen
ProjectVideo Surveillance by Multiple Drones by Deep Learning and Intelligent Vision(2016/11~)
ApplicationEntertainment, Surveillance and security, and Unmanned aerial vehicle
FunctionImage detection, Image recognition, and Object/human tracking
Technical BenefitExpand monitoring scope, Improve automatic level, and Improve system intelligence
Technology StatusCan be transferred

For sports event broadcast, traditionally, large numbers of fixed cameras are spreading. To some large-area sports events, such as marathons or cycling races, due to the limitation of field-of-view of cameras, shooting by photographers or broadcast cars are required but lead to a great cost. Furthermore, such additional broadcast cars may cause the congestion of the traffic in the game and further result in potential danger to players. Due to these circumstances, we proposed an UAV(Unmanned Aerial Vehicle) autopilot system based on character recognition, human detection and human tracking technologies to track sports player automatically. It reduces human labor to decrease the cost, providing a safer environment for players to compete and a better view angle for broadcast. Experiments have shown the proposed system can track human well and is stable even when losing sports players number or facing pedestrian disturbance.

Fig. The concept of the proposed UAV autopilot system for sports player tracking. With computer vision technologies to the images grabbed by UAV, we can track the target athlete, estimate relative positions and orientations between the UAV and the target, and further send control signal to UAV to keep it chasing the target.

demo video: