A Simple Algorithm for Non-cooperative Target Recognition Based on Lidar

Abstract

Aiming at the problem of simple and fast recognition of non-cooperative targets in 3D space, a simple recognition algorithm for point cloud targets is proposed. First, the point cloud data was divided into $n$ categories with the first K-means clustering. Second, the target class was identified with a coarse sieve, and the speed of the algorithm was improved with sparse processing. The more accurate target class was obtained with secondary clustering. The two types of point cloud data are processed by principal component analysis (PCA), which obtains the feature root matrices. Then cosine distance matching was applied to the feature root matrices and target library (trained by 12 groups of point cloud data). This type of data was retained when the similarity was greater than the upper threshold. Therefore, the center point coordinates, distances, and similarity of the target were outputted. The experimental test results of the 13th and 14th groups indicated that the target segmentation similarity of this algorithm could reach 95.75% and 96.98% respectively, and the accuracy reached 100%.