Research On Mobile Robot Algorithm Based On RGBD Camera

SLAM plays a key role in the location and navigation of mobile robots.The visual SLAM refers to the mobile robot using the camera as the main sensor to acquire the image information of the surrounding environment,thereby performing environmental map construction and real-time location.With the rapid development of SLAM technology,it is widely used in augmented reality,unmanned driving and other fields.The most important function about Visual SLAM for mobile robots is location and mapping.Among them,the location is mainly related to visual odometry,back-end optimization and loop closure.The mapping refers to the map building module.The specific research contents of this paper are as follows:First,analyze the research status,future trends and related technologies of visual SLAM at home and abroad.Then introduce the RGBD camera used in this article-Kinect2.0.Zhang Zhengyou's camera calibration principle was studied,and the camera was calibrated under the ROS platform to complete the acquisition and registration of color images and depth images,and the error analysis was perfomed on the calibration results.Second,the visual odometry has been analyzed and researched.The visual SLAM visual odometry is divided into feature method and direct method according to whether the image is extracted.This paper discusses the visual odometry based on feature method in detail.It mainly includes steps of feature extraction,feature matching and motion estimation.The principle of three image features is analyzed,and different feature matching methods are used to analyze and compare different features.The characteristics and matching methods of efficiency and robustness are selected.Then the pose solving algorithm PnP and ICP are studied.Aiming at the distribution characteristics of adjacent image matching features and improving the proportion of correct matching pairs,the accuracy of motion estimation results can be improved.By verifying the effectiveness of the algorithm on the TUM dataset,the error analysis of the experimental results before and after the improvement is carried out.Third,the research analyzes the back-end optimization theory and the loop closure in the visual SLAM,and briefly introduces the different map types and functions in the SLAM.The effectiveness of convolutional neural networks in loop closure is mainly studied.The experimental part on the TUM dataset verifies the role of back-end optimization and loop closure based on the Bag-of-words model in SLAM.The point cloud is spliced according to the solved camera pose,and the experimental results are analyzed.Fourth,the system platform and algorithm flow of the mobile robot are introduced in detail.The complete algorithm verification is carried out on the dataset.Then,the four-wheel omnidirectional CMA20 mobile robot equipped with a laptop and a Kinect2.0 camera is used to complete the point cloud splicing and trajectory drawing of the indoor scene and complete the constructions.And analysis of the resulting experimental error results.Finally,this paper gives a summary and analysis of the research work done.At the same time,the research directions and problems to be solved are proposed.