Based on Robot Positioning and Mapping Decomposition Research on Improving PHD-SLAM Method

Aiming at the phenomenon of high clutter in the scene and a large number of missed detections by sensors, a PHD-SLAM method with untracked Kalman filtering under robot pose building decomposition is proposed. This method avoids complex data association based on stochastic finite set theory. Environmental map estimates are based on estimated vehicle trajectories The first statistical moment of the set propagated by the probability hypothesis density filter effectively reduces the computational burden, and the unscented Kalman filter is introduced in the updating process of the Gaussian term function to improve the accuracy of the SLAM method. Through simulation experiments and KITTI data set verification, compared with the EKFSLAM algorithm, This method avoids the complicated data association process and improves the estimation accuracy of map features and robot attitude. It is worth noting that in the data set verification stage, ORB feature extraction algorithm is used to obtain the feature points in the visual image information and realize the coordinate transformation between continuous image frames. In the subsequent SLMA stage, the position of the feature point after the coordinate change is taken as the measured value, which becomes a new idea to solve the problem of robot positioning and mapping.