A Deep Learning-Based Obstacle Detection System in the Radiotherapy Room based on YOLOv5s

Obstacles left in the radiotherapy room may pose risks to the radiotherapy process. In order to solve the problem of automatic obstacle detection in the radiotherapy room, a system is proposed that integrates an image recognition algorithm for detecting obstacles based on the YOLOv5 neural network. One part involved using a smartphone to capture environmental videos of the radiotherapy room and extract images through frame capture, while the other part involved using a web crawler to obtain manually screened images. The training and testing sets were manually annotated using labelImg software. Then, after setting the parameters, train the YOLOv5 model. Finally, the trained model was tested and the YOLOv5 model performed well in terms of overall performance, with the mAP of the three types of obstacles reaching 0.993, 0.980, and 0.990. To offer users a user-friendly and convenient human-machine software experience, the front-end of the software was developed using PyQt5, achieving the functions of detection and result saving. Through testing, the software has demonstrated accuracy levels comparable to manual obstacle detection, thus establishing its practical value. Consequently, this study presents an intelligent detection tool that is invaluable for researchers involved in safe radiotherapy and intelligent monitoring. Additionally, it can provide valuable guidance for precise monitoring in various settings, including hospital imaging scenarios.