YOLOv10 Improvement Approach Based on ROS Robot Visual Obstacle Avoidance System

YOLOv10 is critical but limited in robot visual obstacle avoidance under market demand. In order to improve the real-time, accuracy and performance in dark light and high noise environments, the study carried out three optimizations: firstly, the quality of low-light images was effectively improved by using SCINet self-calibrating illumination module, which ensures the reliability of ROS obstacle avoidance system at night or under dim conditions. Second, to address the image noise problem of low-bandwidth camera inputs, the study introduce an attention-directed convolutional neural network (ADNet) for efficient noise reduction. Finally, to address the problem of running speed degradation due to increased model size, the study employ a modern inverted residual shift module (iRMB) to build a lightweight EfficientMode. these improvements not only maintain the fast recognition and small size advantages of the YOLOv10 model, but also significantly improve its performance in challenging environments, resulting in improved model mAP values that exceeds 0.596 of the original YOLOv10n, which significantly improves the practicality and efficiency of the ROS visual obstacle avoidance system.