This repository contains a demo of T-RexNet for the specific task of tennis ball tracking. The pretrained network is provided for easy usage. The program takes in input the video of a tennis match and outputs a new video where a bounding box is drawn around the detected tennis ball. The full paper can be found at https://www.mdpi.com/1424-8220/21/4/1252/pdf.
The author used Python, version 3.6.9, and the TensorFlow library,version 1.14. Newer versions may work but it is not guaranteed. The paths below must be included in the interpreter paths: [tensorflowCheckoutPath]/models/research/object_detection/models [tensorflowCheckoutPath]/models/research [tensorflowCheckoutPath]/models/research/object_detection
Just clone this repository and substitute the input video with the video you want to perform tennis ball tracking on. For best performance, given the way T-RexNet works, the camera has to be static.
Please cite this work as Canepa, Alessio, et al. "T-RexNet—A Hardware-Aware Neural Network for Real-Time Detection of Small Moving Objects." Sensors 21.4 (2021): 1252.