A Multi-Modal Portable Guide System Based on The Raspberry Pi Platform

In order to solve the problem of travel safety and security of 1.3 billion visually impaired people in the world, visual perception and obstacle detection are one of the key technologies, and this paper proposes a portable guide system based on improved visual segmentation algorithm and monocular ranging model. The system adopts the Raspberry Pi embedded platform, collects the environment image through the RGB camera, proposes the VE segmentation algorithm based on the mean difference between classes to realize the obstacle contour detection, combines the perspective transformation model to complete the depth calculation, and finally outputs the obstacle avoidance prompt through the speech synthesis module. Experiments show that the system can detect obstacles in the range of 0.2-3m in complex indoor environments, with an average response time of 230ms and a comprehensive accuracy of 92.7%, which is better than the traditional guide stick and ultrasonic guide instrument.