Computer Vision For Aimpoint Tracking In Biathlon

Abstract

The purpose of this thesis was to look at the development of a computer vision system for aimpoint tracking in biathlon. It s desired to develop a system that tracks the Point-of-Aim of the biathlete in order to give quantifiable feedback during and after shooting practice to accelerate skill acquisition. This report covers the process of developing a prototype computer vision system, from the initial idea to a working prototype implementation and system evaluation. It was found that the usage of the Circular Hough Transform (CHT) algorithm gave high (sub-pixel) precision on a wide range of video resolutions and is suitable for future implementations. It came at the cost of expensive computation demands, and reaching high update frequencies were difficult. 13 FPS was seen on the desktop implementation when analyzing a 640x424 video, while 50 Hz were obtained on a 216MHz Cortex M7 analyzing a 20x120 video source. Feature tracking based on establishing a region-ofinterest within the frame was proved to be successful. A successful aimpoint tracking output was obtained and is presented. Camera and processing power specifications are identified and presented for future implementations. The usage of Matlab as a development environment is discouraged for further work on the project, and the switch to a OpenCV-based implementation is recommended.