The project aims to develop a Multi-camera multi-object tracking system to collect information about participants in the keirin cycle race.

Introduction

Three cameras were set up at the circular Keirin race track, with each camera’s field of view covering the starting line, the 50-meter line, and the 150-meter line, respectively. Using the collected race videos, a cross-camera multi-object tracking method was designed to determine the lap count, position, ranking, race time, sprint time, interval velocity, and other relevant information for each participating athlete.

Responsibilities

• Constructed a multi-object tracking model and trained and tested it.
• Researched multi-object tracking algorithms and multitask learning within the JDE (Joint Detection and Embedding) framework.
• Designed a cross-camera matching algorithm.
• Designed a logical method for outputting athlete information based on tracking results.
• Defined the interfaces for web deployment.