YOLO Detection & Automatic Annotation - Unreal Engine Integration
Bringing YOLO-powered object detection to photorealistic virtual and digital twin sports venues
ROLE: Senior Computer Vision & Unreal Engine Engineer / Researcher
SKILLS UTILIZED: Computer Vision YOLO model training, AI scripting (Python), Environment Design & Development, 3D Modeling & Animation (Maya), Data Collection, Data Processing, Data Analysis, Coding (C++, Python)
SOFTWARE: Unreal Engine, YOLO, OpenCV, Substance Painter & Designer, Maya, and TensorFlow
ABOUT: Simulated environments are excellent mediums for visualizing different environmental configurations and designs, as well as simulating sports. However, they also have an incredible untapped potential for reconfiguring, training, testing, and evaluating AI & Computer Vision algorithms / models for real-time athlete performance analysis.
For this project, YOLO, a popular object detection algorithm was integrated with Unreal Engine for the NBA (National Basketball Association). The developed solution allows for customizable object asking by class and automatic bounding box generation in YOLO dataset format. Several system parameters such as bounding box export format, frame capture rate, and masking settings are fully customizable for different sports contexts.