Research On Autonomous Navigation Method Of Lawn Mower Based On Machine Vision

With the continuous development and improvement of machine vision technology,machine vision-based navigation systems have been widely used in agriculture,industry,unmanned driving fields and service industry.Due to the development of urbanization,lawn mowers have become more and more widely used in recent years.However,traditional lawn mowers can only be operated manually,which means intensive labor is need and then results in low work efficiency.With the increase in the speed of embedded processors and the advancement of machine vision technology,it is possible to realize the autonomous navigation of lawn mowers by combining machine vision with embedded systems.Therefore,the study of autonomous navigation and intelligent control of lawn mowers based on machine vision becomes a matter of important theoretical significance and practical application value.This paper studies the machine vision processing method based on the embedded platform and applies it to the autonomous navigation of the lawn mower.The specific research content is divided into the following parts:(1)This paper completes the construction of the hardware platform according to the functional requirements of visual navigation,mainly including the hardware connection of the vision sensor,ultrasonic sensor,and Raspberry Pi 3B+,driver module,and completes the system construction of the hardware platform.(2)The needs to realize the internal and external boundaries recognition in visual navigation considered,different algorithms are proposed here.For external boundary recognition,this paper provides a method based on HSI and CIE Lab color space segmentation and fusion.The improved Otsu algorithm is used to threshold the component images and logical operations are used for fusion,which can effectively deal with the part of the blur under the influence of camera shake or illumination.For the internal boundary recognition,this paper provides a method based on the segmentation and fusion of texture features and color features,using block processing to optimize the extraction of texture features.The texture-based segmentation is completed by SVM and merged with the threshold segmentation based on CIE Lab color space to complete the overall segmentation.The edge detection operator is used to complete the extraction of internal and external boundaries and make corresponding behavior judgments.(3)According to the recognition of obstacles in the grassland required by visual navigation,this paper provides a recognition method based on the combination of color and texture features.It does not distinguish the types of obstacles.From the perspective of the camera,it can select sub-images and obtain color features.All obstacles are classified according to the difference in color features.There are two types of processing,which are respectively based on color feature and texture feature based recognition.Besides,according to the recognition result,the ultrasonic module is activated for real-time ranging and feedback information.The car makes a judgment based on the results of the distance measurement and executes the corresponding command actions.(4)Finally,the experiment to test the above functions with the Raspberry Pi based on Linux system proves that the algorithm researched in this paper has a good recognition effect.