Research on Table Tennis Object Recognition and Detection Based on the Improved YOLOv5 Algorithm

Detecting objects in fast-changing scenes, especially in sports like table tennis, is tricky because targets move quickly and the system must work in real time. This study introduces a simple but effective method to spot ping pong balls using a modified YOLOv5s model, which is perfect for devices with weak hardware. Our approach builds on YOLOv5s-a compact and speedy version of YOLOv5famous for its good balance between speed and accuracy in finding objects. To make it even better, we use tricks like adding more training data by altering existing images, borrowing knowledge from pre-trained models, and cutting unnecessary parts of the network. This helps the model detect balls more precisely without needing too much computing power. For training, we collected many pictures of ping pong balls under different lighting and motion situations, making sure the system works well in actual games. Tests show that our improved YOLOv5s model is both fast and accurate, taking less than 30 ms to find the ball in each frame. Because of this, it’s great for live sports tracking, where every millisecond matters. In short, this research proves that YOLOv5s can be a strong choice for real-time object detection when things move fast, like in ping pong matches.