基于深度学习的盲人行路全方位障碍物检测系统OA

An omni-directional obstacle detection system based on the LPC-YOLO algorithm is developed to address the problems of low detection accuracy and incomplete detection field of view of the existing assisted blind travel navigation system.The image acquisition module of this system consists of cameras distributed in four directions,which are used to collect images around the blind travel path in real time.An improved obstacle detection algorithm LPC-YOLO based on YOLOv8n is proposed.If an obstacle is detected,the monocular ranging algorithm is then used for obstacle ranging.Finally,according to the distance difference,speech synthesis technology is used to broadcast prompt speech for the blind in real time.The LPC-YOLO algorithm improves the SPPF module to reduce the number of parameters and enhance the feature extraction capability.Secondly,a PPA attention mechanism module is introduced to use different sized blocks for multi-scale feature extraction,which improves the performance of small-target detection.Lastly,a CAFusion feature fusion module is proposed to extract the low-level features and high-level features and perform feature fusion to further enhance the feature extraction capability.The experimental results show that the improved LPC-YOLO model improves the mean accuracy(mAP50)by1.5 percentage points over the original YOLOv8n model on the obstacle data set.The system also performs well in terms of obstacle detection and voice prompts in tests under sunny weather conditions on outdoor roads,which can provide effective travel assistance to the visually impaired.