Research On Obstacle Detection And Recognition Method Of Snow Robot

Obstacle detection and recognition has been a hot topic in the field of mobile robot,and it is also the first prerequisite for robot safety operation.This paper starts with the requirement of the obstacle detection and recognition in the design of the practical application project of snow robot,firstly designing an obstacle detection method based on the principle of ultrasonic distance measurement,and then making full use of deep learning learning image features from big data automatically,discussing the application of deep learning algorithm in robot obstacle detection and recognition.It provides a new method and idea to solve the robot obstacle detection.The main contents include:Designing a robot obstacle detection system based on the principle of ultrasonic distance measurement.On the basis of understanding the basic principle of ultrasonic ranging and the characteristics of the ultrasonic sensor,designing an obstacle detection system for distance measurement.From the obstacle detection experiments,finding two factors that affect the accuracy of ranging,which are called probe aftershocks interference signals and the strong interference signal suddenly,influencing the results.Exploring the feasible optimization method to reduce the influence of probe aftershocks interference signals and the strong interference signal suddenly as far as possible,and design comparative experiments to verify the feasibility of the optimization method.The optimized ultrasonic distance measuring system has good anti-jamming performance,and it can detect the distance of the front obstacles more accurately and quickly.Designing a robot obstacle recognition system based on deep learning,and the specific categories and location information of obstacles are identified through deep learning algorithm.The Kinect depth camera is used to realize the ranging work.Making the data set suitable for the target detection algorithm to train the original target detection algorithm.The network structure is optimized by using the MFM activation function and the network structure of NIN in the basic structure of Single Shot multiBox Detector.Through the comparison of the results before and after optimization,it is found that the accuracy of the test is improved by 7.4%,which highlights the effectiveness of the optimization method.The color image is photographed with the RGB camera of the Kinect depth camera,and the infrared camera takes the depth image.Using the optimized target detection module to get the coordinate information of the object in the picture,and combining the depth image to get the average value of the pixel distance in the specified range of the depth image.The average value is used as the distance between the object and the camera.Using the mode of target detection and depth measuring to simulate the snow robot's identification and location on the front obstacle.