Twitter user measure_plan has developed a computer vision system that analyzes his tennis training and tells him what to work on. To do this, just record the lesson on your iPhone.
The system determines the speed of the ball, distinguishes between forehand and backhand hits, and tracks where the ball lands and the position of the player’s body at the moment of impact. All results can be seen directly on the marked video.
For processing, the developer uses a CV pipeline based on Roboflow Agent: data tagging is automated by Astra 6, object detection is performed by a pre-trained RF-DETR model, and the final video tagging is done using Python and OpenCV.
The project is still at an early stage. Measure_plan wants to turn it into an app that will record every workout, track progress over time, and give recommendations for improving technique.
The first training sessions have already shown the practical benefits of the system: after looking at his own recordings, the programmer realized that he needed to work more seriously on his leg movement and backhand.
MLB banned the use of AI during matches – neural networks suggested tactics to coaches
In the comments, however, he was reminded that numbers like ball speed and number of shots alone are not enough for progress. One suggested focusing on hip work and striking technique, while another suggested add two indicators: the distance from the player to the ball at the moment of impact and how early the player takes the correct position after the rebound.
Measure_plan is already going to add an “ideal strike zone” to the system and compare the player’s position relative to it. In the future, this will make it possible to evaluate not only the shot itself, but also how correctly the tennis player managed to reach the ball.
