Study Of 3D Mapping Method Based On Lidar

Simultaneous localization and mapping(SLAM)have become very wide and continuously expanding area of research since last decade.Realizing the number of its applications which ranges from precise indoor mapping to rescue operations,researchers are more focusing on the precision that SLAM brings to table due to its highly anticipated application in autonomous driving vehicles.These days SLAM algorithm are capable of utilizing multiple sensor with number of sensor being directly proportional to the precision in mapping and localization.Precise mapping is one of most important aspect of not only autonomous driving vehicle but also for applications where mapping is only focus,i.e.cave exploration,human rescue operations etc.Lidar is accurate sensor widely used for this purpose but have its limitations in terms of field of view.Actuating a 2D Lidar in 6-DOF results in providing 360-degree field of view but also falls a victim to motion distortions in point clouds due to relative linear and angular motions.Techniques developed before to overcome these errors are mostly offline which tends to correct the registration of point cloud by implementing loop closure over time.In order to address this problem in real-time,usage of redundant sensors like camera is essential.Focus of research in this dissertation is eliminating motion distortions in point cloud by improving the scan registration utilizing IMU data.This technique results in detailed and precise mapping of complex environments while remaining low-cost solution.To make the Lidar motions precise,stepper motor controlled by Arduino is used eliminating the needs of extra angle sensor.A point-to-plane iterative closest point algorithm is implemented to ensure precise point cloud registration.Registered points are categorized into edge and planar point decided by the odometry algorithm.A motion estimation sub-algorithm within odometry accounts for the time stamps,angle of sensor and IMU information to calculate the relative distance information of edge and planar point and their relative correspondences which lessen the possibilities of false registration.Feature points area also turning map into easily readable by human.Mapping and odometry algorithms are run simultaneously for this method to be a real-time method.Prior algorithm is designed to run at a much lower frequency for fine results than posterior.However,IMU unit is still kept optional during this implementation Results acquired from this technique corroborate substantial improvements in mapping and feature extraction of schemed method when used with the IMU.The aim of accurate mapping solution while using a low-cost sensor is achieved with future adaptations to hardware and algorithms in consideration.