Research On Closed-loop Detection Method Of SLAM Based On Particle Swarm Optimization

Compared with traditional laser and infrared sensors,visual sensors can provide useful information with more details but lower cost.Simultaneous localization and mapping(SLAM)technique based on graph optimization has always been one of the research hotspots.In the visual SLAM technology,calculating the pose of a robot by the data association information of the images of adjacent frames will make the accumulated error of the previous moment inevitably be transmitted to the latter moment,resulting in the drift of the robot pose,and ultimately unable to construct a global uniform trajectories and maps.Introducing the closed-loop detection module will add an additional constraint to the pose of the robot,which can largely eliminate the cumulative error of the robot's pose,resulting in a globally consistent trajectory and map.In this paper,the closed-loop detection module of SLAM technology is studied in-depth.For current closed-loop detection based on the bag of words model,there is the time-consuming feature of off-line training,the dictionary is slow to load,and the data association information is not taken into account when representing images with discrete words,due to the possibility of closed-loop misjudgment,a closed-loop SLAM detection method based on particle swarm optimization(PSO)is proposed,and a "histogram reduction dimension"strategy is proposed to compress key image information.The robust descriptor of the image is obtained as an optimization model of the optimizing particle swarm algorithm,further improves the convergence speed and reliability of the algorithm.The main works of this paper are as follows:(1)In-depth study of the imaging principle and data acquisition process of the RGB-D sensor.Through the calibration of the color camera and the depth camera,the internal and external parametric model of the Kinect camera is obtained to provide accurate images for subsequent image feature matching and pose estimation.(2)Propose an image robust descriptor extraction method.A "histogram reduction dimension" strategy is proposed to reduce the dimensionality of the binary string of BRIEF descriptors of the ORB feature extraction algorithm.The robust descriptor vector of the image is obtained,which shortens the feature extraction time and effectively avoids the lack of data association when discrete words are used to represent the image.Experimental results show that the extracted image robust descriptor vector can effectively represent the image,not only saves storage space but also has high operating efficiency.It can meet the real-time operation requirements of SLAM in the closed-loop detection process.(3)This paper proposes a SLAM closed,loop detection method based on particle swarm optimization algorithm.The particle swarm optimization algorithm is applied to the closed-loop detection,and the particle swarm optimization algorithm is used to automatically find the image with the highest degree of matching with the current frame image in the historical frame image.By introducing the constraint of image fitness threshold,the accuracy and recall rate of the closed-loop detection method can be evaluated.Simulation experiments show that compared with the closed-loop detection method based on the bag of words model,the closed-loop detection method proposed in this paper takes less time and has higher accuracy.