AI-Driven Scouting Application (ADSA) uses a custom data set of 500 images and more than 3350 bounding box annotations to build a custom YOLOv8 model. This paper presents a comprehensive overview of real-time object detection models to detect and classify important objects in video streams of the FIRST Robotics Competition (FRC). In addition, this article elaborates on related work regarding object detection for scouting in sports, other types of ranking systems used in FRC such as EPA and OPR, analysis of ADSA’s performance on specific types of data, and lastly, future work that might be added to ADSA. Overall, ADSA has precision (P) and recall (R) values at 0.822 and 0.781 respectively, and performs exceptionally well in AMP and speaker classification.
This paper can be accessed here
