Research On Scene 3D Reconstruction Technology Based On Li Dar And IMU

Three-dimensional reconstruction technology is widely used in unmanned driving,surveying and mapping,logistics,medical and other fields.Three-dimensional scenes provide reliable a priori information for autonomous decision-making by machines,and are the prerequisite for machine intelligence.The 3D reconstruction technology uses environment-sensing sensors to register sensor data at multiple moments in a unified coordinate system to achieve data fusion and form an offline 3D scene.Improving the registration accuracy is a long-term research goal pursued in this technical field.In order to solve the problem of sparse scanning points of low-line-count lidar,dynamic scanning is used to obtain rich scene information of multi-frame point clouds.Aiming at point cloud distortion and point cloud scene ghosting,this paper introduces an inertial measurement unit(IMU)into the lidar measurement system to study the point cloud registration method of lidar point cloud and pose fusion.A 3D reconstruction test system was constructed using ROS(Robot operating system)platform,C++,PCL library and Cloud Compare software.Aiming at the problems of mismatching points,long registration time and low registration accuracy during the registration process of two-frame point clouds,a three-dimensional scale invariant feature transformation(3DSIFT)key points and a binary direction histogram(BSHOT)descriptor are proposed.Combining the method of constructing point cloud feature matching pairs.In the indoor experimental scenario,the registration accuracy is improved compared with the commonly used methods,and the calculation time is reduced at the same time.Aiming at the point cloud distortion in radar motion,a point cloud de-distortion method based on IMU linear interpolation is proposed.Align the point cloud data with the IMU pre-integrated pose in time to obtain the pose matrix of the radar,and use the matrix to correct each laser point in a single frame of point cloud to the coordinate system where the starting laser point is located.Aiming at the ghost phenomenon of point cloud scenes caused by accumulated errors,a multi-frame point cloud registration method that introduces IMU radar pose information is proposed.Taking the IMU radar pose as the initial registration pose,a joint optimization function including the point cloud registration error and the IMU pre-integration error is constructed.Experimental results show that the introduction of IMU improves the accuracy of multi-frame point cloud registration.