A Real-Time Monocular Vision-Based Obstacle Detection

In recent years, many robotic obstacle avoidance methods use high-priced laser detection technology, or a dual-lens camera as a sensor for robot control. In order to reduce costs, this paper presents a real-time obstacle detection system by using a single wide-angle camera. In the robot application, the system can detect various static or dynamic objects in front of the robot body. The proposed system uses a local object based background subtraction method to get the outline of the static/moving object. The adaptive integral image based obstacle detection method without manual adjustment of setting parameters. The obstacle detection system had been implemented by ARM system, the obstacle can be detected on the real-time camera. The frame rate is close to 32 fps with Raspberry pi 3 platform.