Object tracking for sports analysis (Part A)

• Main Author: Wang, Zhihao
• Other Authors: Andy Khong Wai Hoong
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/67731
• Institution: Nanyang Technological University

This report includes details of sports analysis by implementing detection and tracking of a single athlete, given his training or match video footage taken by stationary cameras. The video footage will be processed in a frame by frame manner by utilizing FFmpeg tools. And the detection and tracking procedures can be conducted in an internet-free environment. A user prompt calibration step is required in the first place for the matlab program to capture the initial frame without any foreground tracked object as background image throughout the analysis. Following that, some minimum and maximum values are determined and recorded for each pixel in the calibrated background image after transforming it from RBG to grayscale. Upon the tracked object is introduced in the following video frames, if some pixel values in each consecutive frame have fallen outside the preset minimum and maximum value ranges, it is an implication that the tracked object might have moved into those pixels. Knowing the position of the object within each video frame, calculation on displacement that the object has moved between each video frame can be performed by Euclidean Distance Formula, taking the object’s centroid coordinates in the current and previous frames as the four formula parameters. Eventually, the object’s total travelled distance can be obtained by summing up all displacements between each frame. Moreover, information such as the highest moving speed and the travelled paths (heat map) of the athlete will also be available in the Matlab output for better physical observation and performance analysis.