A Method for Obstacle Detection in Radiotherapy Room Based on Improved YOLOv5

The presence of obstacles in the radiotherapy room can disrupt equipment operation and even lead to accidents. To address this, monitoring cameras could be installed to detect these obstacles. In this study, an intelligent obstacle recognition model for radiotherapy rooms is proposed. The model utilizes deep learning to automatically extract obstacle features, eliminating the need for human monitoring and improving the quality and speed of obstacle detection. By proposing the improved YOLOv5 model, which replaces the original module with MobileNetV3 in the backbone section of YOLOv5 and utilizes CA attention mechanism instead of the SENet attention mechanism of MobileNetV3, the detection model’s parameter count is reduced, and the ability to capture spatial information with low sensitivity is enhanced. To evaluate the model, a self-made dataset consisting of commonly encountered obstacles in radiotherapy rooms. Evaluation indicates that while ensuring detection accuracy, the proposed model reduces the parameter count and weight file size to 3187527 and 6.7MB, respectively, which are 3840193 and 7.7MB lower than the unmodified YOLOv5, and 360623 and 0.7MB lower than the MobileNet-YOLOv5. The results demonstrate that the detection efficiency of obstacle recognition in radiotherapy rooms can be significantly improved by applying improved YOLOv5 model.