Smart Phone Based Visual Obstacle Avoidance System For Mobile Robot

Navigation is a vital research branch of robotics, in which obstacle avoidance is a primary component. In recent years, the evolution of mobile computing technology, which is represented by smart phone, provides a commonly used computing platform for the application of machine vision. This paper combines robotics with machine vision and mobile computing to develop and implement an algorithm for outdoor obstacle avoidance in real-time based on monocular vision.The system we develop uses the camera of a smart phone to capture pictures, and a monocular vision algorithm is applied to obtain the position of obstacles. The algorithm has several hypothesis:all objects in the scene are perpendicular to the ground and the ground region is in the bottom of the image. Obvious edges that are locally continuous and smooth exist on the common boundary between ground region and other regions due to the inconsistency of their color and texture. Based on these hypothesis, an edge detector can be used to get the border between the ground region and objects perpendicular to the ground. In order to remove the influence of shadow, image fusion is conducted in the HSI color model for each channel before obtaining the border.When locally continuous ground border is obtained, it is fitted by improved ACM to get a smooth ground border, and the depth of the point on that border indicates the distance between the car and obstacle. The robot will always choose the point on the ground border with a biggest distance from the car as the direction to move forward in order to avoid obstacles.The robot used in this thesis is a refitted car equipped with a controller and sensors. To fulfil the control of it, corresponding sensor and controller are added. Specifically, the car posture is obtained using the inertial sensor of the smart phone to control the rotation of the car. Besides, a PID controller is designed using the car speed form the closed-loop feedback control system and the performance is improved. The experiment results show that the car can avoid obstacles correctly even when it drives with a high speed.