Multimodal Perception System for Real Open Environment

This paper presents a novel multimodal perception system for a real open environment. The proposed system includes an embedded computation platform, cameras, ultrasonic sensors, GPS, and IMU devices. Unlike the traditional frameworks, our system integrates multiple sensors with advanced computer vision algorithms to help users walk outside reliably. The system can efficiently complete various tasks, including navigating to specific locations, passing through obstacle regions, and crossing intersections. Specifically, we also use ultrasonic sensors and depth cameras to enhance obstacle avoidance performance. The path planning module is designed to find the locally optimal route based on various feedback and the user's current state. To evaluate the performance of the proposed system, we design several experiments under different scenarios. The results show that the system can help users walk efficiently and independently in complex situations.