A Multi-camera Optical Flow Vision Odometry Based Incremental Robot Localization Method

Mobile robot localization is one of the core problems of robot motion,and the traditional odometry localization had accumulated localization errors during its motion.Starting from the vision odometry localization method,this paper explores multi-camera vision odometry.The traditional multi-camera vision localization algorithm is to sense the surrounding environment by camera,but it is easily disturbed by the environment.In order to reduce the interference of the environment,this paper proposes a three-camera optical flow localization algorithm.A robot platform based on multi-camera three cameras position of the visual odometer is designed,and the three cameras of the robot are placed at three position points with equal height level on the robot platform.According to the Cayley-Menger's triangular gravity coordinate formula,the current position of the robot can be calculated from the conformational relationship between the three cameras and the robot positioning points.The camera acquires ground images for real-time localization,followed by calculating the optical flow velocity based on the LK pyramid optical flow algorithm.And the motion velocity of the camera is calculated based on the spatial position transformation of the feature points and the actual points.And the velocity of the camera in the camera coordinate system is transformed into the velocity of the robot's world coordinate system based on the three cameras position fusion of three-camera velocity,and the position pose of the robot is calculated by the trajectory calculation.Due to the limitations of camera perception of the surrounding environment in the visual odometry.The camera collecting ground image capture method reduces the chance of environmental interference.Then analyze the nonlinearized calculation method based on the expectation and covariance of the positioning error of three cameras.And the positioning error of different triangular configurations is simulated and analyzed.It is proved that the error is minimized when the positioning point is located at the area equipartition point of three cameras,and the positioning accuracy of three-camera fusion is better than that dual camera and single camera positioning.With the analysis of robot localization errors under different trajectories and different number of cameras,simulation verifies the advantages of three camera localization over dual camera and single camera,and the optical flow odometry based on the three-camera algorithm is better than wheeled odometry in localization position and angle accuracy at different trajectories and sampling frequencies.The applicability and accuracy of the algorithm with different trajectory movements are demonstrated.The three-camera optical flow robot system is designed and experimentally tested to make the robot do rotation,translation,and rectangular motion test in a plane,and the fused position information is obtained by collecting the optical flow velocity through the robot's three cameras.The high sampling frequency of the cameras and making real-time positioning position curves demonstrate the accuracy of the position algorithm.This paper implements incremental robot localization with the three camera optical flow position algorithm can satisfy the low-cost condition with high localization accuracy,real-time and robustness.