Research On SLAM System Based On Stereo Camera

SLAM(Simultaneous localization and mapping)has a wide range of applications in the areas of driverless,robotic,and AR.How to accurately position camera poses and complete high-quality mapping has always been a problem that scholars have been ex-ploring.At present,most SLAM systems mainly use monocular cameras.Monocular cameras are relatively inexpensive and easy to obtain.Monocular cameras can provide very rich scene information,but monocular SLAM systems cannot estimate the true scale of the scene.Although the lidar-based SLAM system can quickly acquire depth information,relying only on the point cloud information cannot complete reliable loop closing detection,and it is difficult to eliminate accumulated errors,and the price of lidar is too high.Multi-sensor fusion SLAM systems require complex hardware setup and synchronization calibration.The stereo SLAM system requires only a stereo camera,which not only makes full use of visual information,but also accurately obtains depth information,and solves the problem of inconsistent scales of monocular SLAM.We design a simple and efficient stereo SLAM system.The SLAM system is quickly initialized according to the stereoscopic feature points,feature matching is performed on continuous frame views,data association is completed,camera pose is estimated,and it is optimized.As the camera moves,new views are evaluated,and the key frame is selected.According to the key frame information,carries on the loop closing to the whole SLAM system,reduces the accumulated error,and completes drawing the entire scene map.We have further optimized camera poses and 3D spatial points.In the Bundle Adjustment,we further eliminated the H matrix based on its sparseness,prioritized changes in the camera pose,and then optimized the 3D space points to make full use of the The characteristics of the corner block matrix speed up the back-end optimization.In order to evaluate the effect of the stereo SLAM system,we proposed a method for evaluating camera pose using a checkerboard.In order to more intuitively express the robustness of the binocular SLAM system,we added the AR module to reflect more intuitively.Experiments show that our stereo SLAM system can stably estimate the camera pose and complete the mapping of the space map.