3D Dense Map Reconstruction Based On Visual SLAM

Simultaneous Localization and Mapping(SLAM)technology enables robots to localize themselves and estimate their posture in unknown environments,and build maps of unknown environments.With the advantages of high accuracy,low cost and wide applicability,visual SLAM technology can be applied to a wide range of scenarios and fields.However,the complex and variable lighting conditions lead to a large randomness of extracted feature points and uncertainty.When building maps in real-time,feature points need to be constantly updated,resulting in redundant feature points and reduced map building speed;feature points are too dense and overlap,increasing the number of mismatched features.The feature point method is sparse and contains little information about the environment,which makes the robot’s understanding of the scene poor.To address the problem of large randomness in feature point extraction,this paper constructs an adaptive threshold expression based on factors such as illumination and contrast and proposes an adaptive threshold fast feature point extraction and description(Oriented FAST and Rotated BRIEF,ORB)algorithm to ensure the certainty of the number of extracted feature points.In this paper,a uniform quadtree method is used to homogenize the feature points,and a random sampling consistency algorithm is combined to eliminate the mismatched feature points in order to improve the speed and accuracy of map building.In this paper,a dense point cloud system is introduced to build a dense point cloud map for accurate localization navigation and obstacle avoidance.For variable lighting environments such as darkness,over-brightness or dynamic illumination,the number of feature points extracted using existing algorithms is highly variable.In this paper,an adaptive threshold T expression is constructed based on factors such as illumination and contrast,and an adaptive threshold ORB feature point extraction algorithm is proposed.Under different lighting conditions,the number of feature points extracted using this algorithm remains stable and contains richer image information.Experiments show that the number of feature points and extraction speed of feature points extracted using the adaptive thresholding ORB algorithm under different data sets and different contrast conditions are both improved compared to the ORB algorithm,and the adaptability to illumination is significantly improved.When performing real-time map construction,the tracking module continuously updates the feature points,resulting in a pile-up of feature points,which in turn leads to poor mapbuilding speed.In addition,the camera pose estimation error is high due to feature point mismatch.In this paper,a quadtree uniform distribution of feature points is introduced in the adaptive thresholding ORB algorithm to homogenize the feature points and reduce the feature point buildup.A random sampling consistency algorithm is used to eliminate the mismatched feature points to improve the matching efficiency.Experiments show that the proposed ORB algorithm with adaptive thresholding improves the correct matching rate by 12.9% on average and reduces the extraction time by 16.3% on average.For the constructed map information consisting of point features and line features,there are isolated points and disconnected lines,which are prone to missing environmental information,making the robot poor in understanding the surrounding environment and unable to complete accurate environment reconstruction.In order to improve this situation,this paper introduces a dense point cloud module in the adaptive thresholding ORB-SLAM2,which contains more 3D information,and constructs a dense point cloud map through point cloud alignment and fusion to achieve 3D map reconstruction.The experiments show that the absolute trajectory error and relative trajectory error are reduced by 9.7% and 7% respectively compared with the ORB-SLAM2 algorithm,which effectively improves the trajectory error and robustness of the map,while the adaptive thresholding algorithm can complete the reconstruction of the surrounding environment in real time to ensure that the robot can complete accurate positioning and navigation in the unknown environment.