Research On Tire Recognition And Location Method Based On Monocular Vision

With the popularity of cars and airplanes in China,cars and airplanes have become the necessary means of transportation,and tires as their necessary components,if they are not repaired and replaced in time,it is very easy to cause safety accidents.At present,the traditional manual method is mainly used to repair and disassemble the tire,which costs manpower and material resources and is extremely inconvenient.At the same time,with the rapid development of vision technology,it is possible for vision based automatic tire repair and disassembly.However,at present,only vision technology is used to detect tire defects and trademarks.Therefore,this paper proposes a tire recognition and positioning method based on monocular vision,which lays a foundation for automatic tire repair and disassembly in the future.The main contents and achievements of this paper are as follows:1)The data set of training model is made.All the data sets needed in the experiment were obtained from Internet and real-world photography.A total of 1927 images were obtained from the data set.Because of the noise and interference background in the obtained tire image,it is not conducive to the later feature extraction,so the image preprocessing must be carried out.By selecting various image preprocessing methods for comparative experiments,the weighted average grayscale method,bilateral filtering method and Grabcut image segmentation method are selected to preprocess the tire image.2)The tire identification model is built.In the environment of Py Charm compilation,the tire recognition models of support vector machine(SVM),support vector machine combined with histogram of oriented gradient(SVM-HOG)and support vector machine combined with local binary pattern(SVM-LBP)are built.The above models are tested on the training set and verification set respectively,and the recognition accuracy is 88.91%,56.13% and 85.33% respectively The recognition accuracy of SVM model has been very high,but it still can’t meet the accuracy of industrial tire recognition,so this paper designs the network model of Res Net-20 according to the residual network(Res Net)structure,and the recognition accuracy of this method is 97.4% through the experiment.The experiment shows that the recognition accuracy of Res Net-20 model is high,the structure is stable,and the accurate recognition of tire is realized,so Finally,this method is selected as the tire recognition method.3)The tire positioning method is designed.This paper uses Zhang Zhengyou’s camera calibration method to calibrate Logitech C270 camera,obtains its distortion parameters,builds a positioning model on the basis of Res Net-20 model,when the tire is identified,retains the tire image and corrects the deformity,uses Hough transform to detect the circle,extracts the pixel value of its center and radius,At the same time,according to the known parameters of the camera and the coordinate relationship between the camera and the tire,the tire positioning algorithm is designed to realize the tire positioning.By comparing with the actual value measured by the actual laser rangefinder,the relative error of the distance and angle of the positioning method is about 3.08% and 3.14% respectively.The results show that the method is reliable and feasible for tire positioning,which provides the possibility for automatic maintenance and disassembly in the later stage.