| Automated Guided Vehicles(AGVs) are widely applied to large enterprises at home and abroad.What is more,its obstacle avoidance has been paid more and more attention.However,there are a lot of problems remain to be researched and solved.This paper makes some researches and analysis on the path planning of AGV based on Kinect sensor,focusing on the obstacle identification and avoidance.According to the obstacle detection algorithm,the depth image from Kinect is processed.Because Kinect sensor will produce "speckle noise",the error in the depth of the image is caused.Meanwhile,it does not need very complicated processing algorithm.Therefore,average filtering algorithm is used to eliminate noise.OTSU algorithm is used to extract the foreground region from the depth image by selecting a threshold to maximize the diversity of the variance of the gray value between the background and the foreground.The method has the advantages of low computational complexity and strong self-adaptability.OTSU algorithm mistakes part of the road for the foreground,so U-V parallax method is used to distinguish between roads and obstacles through the plane problem being changed into a straight line problem.In the V-disparity map,the least squares method is used to extract the straight lines in order to determine the intersection of the obstacle and the road and the height of the obstacle.In the U parallax map,the straight line of the obstacle is extracted so that its width is determined.Thus,the specific coordinates of the obstacle information can be obtained accurately.When obstacle is identified,the obstacle avoidance algorithm based on fuzzy neural network is designed.The algorithm has the characteristics of prior knowledge,logical reasoning and self-learning.The depth information after the pretreatment as the input is transmitted to the controller of the obstacle avoidance system,and then the membership function and the fuzzy rule are trained by the learning algorithm,so that the control precision of the system has been greatly enhanced.Aiming at the slow convergence rate of learning algorithm for fuzzy neural networks,a learning rate decreasing method is designed,which greatly improves the convergence rate of the algorithm. |
