Mobile Robot Local Positioning Algorith Based On Visual Information Fusion From Front And Rear Cameras

With the rise of driverless technology,mobile robots autonomous driving has received more and more attention,and it related technologies have also been extensively studied.The autonomous positioning is the key technology for mobile robots to realize autonomous movement,and it is also the first problem to be solved.Because the camera has the advantages of low price and no zero drift error,and with the improvement of computer processing ability,robot positioning by using camera becomes one of the important research of robot autonomous positioning.In complex scenes,the positioning algorithm using camera will be affected,because the scene texture may be sparse and the feature points may be scarce,so there is always to increased cameras to obtain more environmental information.Therefore,this paper mainly studies the algorithm of local positioning of mobile robots by front and rears cameras.We established the coordinate system of the world coordinate system and the camera coordinate system,according to the robot vision positioning system,and analyzed the imaging model of the depth camera,and calibrated the internal and external parameters of the depth camera to complete the transformation relationship between the coordinate systems.Image registration between color images and depth images is achieved based on internal and external parameters of the depth camera.And we build the physics simulation model in the Gazebo simulation software in the ROS system to obtain the image data of the robot during the movement and its own motion data.Designing a visual odometer using the image information of front and rear cameras to achieve local positioning of the robot.The ORB feature points are used to realize the data association in two consecutive frames.For the characteristics that the ORB feature points are easy to be concentrated,the method of dividing the image into several sub-regions and extracting the feature points is adopted to improve the feature point dispersion.In the feature point matching process,the combination of the minimum Hamming distance and the RANSAC algorithm is used to effectively improve the correctness of feature point matching.We used matching feature points of one camera image to design the visual odometer,and used matching feature points of front and rear cameras image to design the visual odometer.Through simulation verification,it shows that the visual odometer coordinated by the front and rear cameras has higher positioning accuracy and robustness.Optimize the local positioning result of the visual odometer according to the characteristics that in the front camera will reappear in the rear camera with the robot moves.For the same scene,the direct feature point matching error between the front camera image and the rear camera image is large.So,we combine the template matching and image re-projection to match the feature points,and obtain the positioning between the current state and the historical state.Using the odometer results and the direct estimation results of the feature points,we established the optimization function,and use nonlinear optimization algorithm to optimization the function.Experimental results show that the algorithm can improve positioning accuracy...