Research On Object Detection And Localization Technology Based On RGB-D Camera

Object detecton and localization technology base on machine vision is widely applied in various tasks.Vison algorithms are used to process and analyze the images acquired by the camera to achieve automatic detection and localization of target objects,and then the posture information is fed back to the execution machine to help the machine realize various intelligent behaviors,such as cargo sorting,automatic assembly,mobile robot navigation,and automatic driving.Researchers have proposed many solutions to the above application scenarios.According to the different types of vision sensors,it can be roughly divided into2 D image-based solutions and 3D image-based solutions.2D images can be fast processed though lack depth information.In particular,deep learning,which shines in the field of computer vision,has greatly improved the efficiency and robustness of 2D image processing when compared to traditional algorithms.The 3D image records the spatial coordinate information of the three-dimensional scene relative to the sensor itself.The information is direct and accurate though the processing speed is slow.As RGB-D cameras can output RGB image and depth image at the same time,they are widely used in various visual solutions.This research project takes the segment grabbing task during the shield construction as the research background,and aims to realize the automatic detection and posture measurement of segment bolts(grabbing components fixedly mounted on the segment)through machine vision technology.This paper,which is mainly focused on object detection and localization technology based on a RGB-D camera,proposes a object detection and localization algorithm.The effectiveness,robustness and positioning accuracy of the algorithm are verified through experiments.The research content of this article is divided into six chapters,which are summarized as follows.Chapter 1 introduced the research background and significance of this dissertation,as well as the domestic and foreign research overview of object detection and localization technology.The research objective was proposed and the research content was determined in this chapter.In addition,the research goal and content,as well as the the route of the research are also summarized in this chapter.Chapter 2 introduced the principle of camera imaging and depth measument.This chapter has also shown how to realize camara calibration according Zhang Zhengyou's calibration method and how to find out the transformation relationship between the RGB camera coordinate system and the depth camera coordinate system.Chapter 3 shown the details as well as the improvements of YOLOv3 algorithm.An improved non-maximum suppression strategy is proposed for segment bolt positioning and positioning tasks in similar industrial scenarios.The network model training was completed and has been tested on the test data set.Chapter 4 completed the construction and screening of the point cloud of the target area.The posture of the target object are calculated based on the improved SAC-IA algorithm and the ICP algorithm.In chapter 5,the object detection and localization program is developed.The performance of the algorithm described in this paper is verified under no-interference conditions,medium-interference conditions,and strong interference conditions.Furthermore,Chpater 5 has also carries out experiments on spatial positioning accuracy.In chapter 6,the research contents of this dissertation were summarized and the shortcomings of the existing work were analyzed.Meanwhile,the future research work as well as the directions for further exploration and research are prospected in this chapter.