Lidar Sensor-Based Object Recognition Using Machine Learning

Light detection and ranging (LIDAR) sensor provides complicated and large volume of environmental point cloud data that are essential for the target recognition. In this study, we present a novel LIDAR sensor-based target recognition approach of the point cloud data, using adaptive rounding algorithm and optimized support vector (SVM). First, the multilayer LIDAR is used to obtain three-dimensional point cloud data of the surrounding environment. Second, we use the grid-occupied method to cluster the point cloud data after reducing redundant point cloud data through a self-adaptive removal method. Third, the multiple features of target are extracted, which classified by a novel SVM based on block feature. Finally, the classifier achieves the best effect through parameter optimization, and the different target objects are distinguished. The experimental results show that the classification accuracy of the target recognition method proposed can reach 93.75% under the premise of reducing training features by at least 33.25%. Therefore, the target recognition method proposed in this study can complete the classification of objects in the target area with a significant increase in accuracy compared to previous approaches.