Improved Template Matching Based Stereo Vision Sparse 3D Reconstruction Algorithm

Stereo vision 3D reconstruction using keypoints matching methods can bring up problems, such as uneven distribution of points, or even mismatches. To solve the problems, we proposed a novel stereo vision 3D reconstruction algorithm based on template matching. First, considering that the similarity between left and right image of binocular camera can be high, template matching can be brought in to minify the searching range of every keypoint, thus enhance the accuracy of matches and make the reconstruction result more complete. The specific process of the proposed method includes the following steps: templates extraction, keypoint density index calculation, template matching, keypoint matching within template matched areas, point cloud generation, and point cloud stitching. Furthermore, a simulation environment was built up in Gazebo, and 3D reconstruction was carried out using the proposed method. Comparison experiment was carried out using brute force keypoint matching method and FLANN. Simulation results showed that the proposed method performed better and improved in both utilization of keypoints and reconstruction result.