Indoor Obstacle Avoidance Of NAO Robot Based On Monocular Vision

In the process of humanoid robot walking,it is extremely important to realize obstacle avoidance behavior,which can avoid the damage of the robot itself and the damage of important internal equipment,so that the humanoid robot can operate normally under any circumstances.For the purpose of enabling the robot to better perceive the external world,a variety of sensors are installed on the robot,and visual sensors are an important part of them.Compared with other sensors,visual sensors are cheap,easy to install,wide range of detection and rich in information.Therefore,the combination of monocular camera and robot vision is a better way to avoid obstacles.Based on monocular vision,this thesis studies the obstacle segmentation,location and obstacle avoidance behavior of NAO robot in indoor unlknown environment.The main work of this thesis is as follows;1.A two-dimensional Otsu speed optimization and threshold optimization algorithm combined with fruit fly algorithm is proposed.Compared with other threshold seg-mentation algorithms,two-dimensional Otsu algorithm has better anti-noise ability but poor real-time performance.Therefore,the speed of the algorithm is improved when the optimal threshold is found by combining the fast search ability of the fruit fly algorithm.In addition,this thesis proposes a threshold optimization method to solve the binarization problem of unclear noise separation and uneven illumination in the two-dimensional Otsu algorithm.After uniform illumination and mean filtering,the optimal threshold value of obstacle segmentation is obtained according to the threshold value calculated by fruit fly algorithm and the characteristics of gray histogram of image.2.A monocular vision ranging model is proposed.Based on the relationship between the angle ratio and the length ratio between the image and the real world,the forward distance and the lateral distance between the obstacle and the robot are derived,and then the real distance and the angle between them are obtained.This model is independent of the height of the obstacle and only depends on the position of the feature points in the image.Moreover,it can complete the distance measurement under the condition that the internal parameters of the camera are unknown.3.A monocular vision obstacle avoidance algorithm based on obstacle detection area is proposed.According to the image segmentation and position information of the obstacle,the appropriate angle deflection strategy is selected in the obstacle detection area to avoid the obstacle in real time in the indoor unknown environment.This will play a positive role in promoting the robot to integrate into human life and achieve the goal of NAO robot to move to the target position smoothly in the actual indoor environment.