Research and Engineering Implementation of Laser Point Cloud 3D Target Recognition Algorithm in Complex Scene
Shixi Shi,Yifu Zhou,Yang Sun,Ziquan Ding,Xiaoxiang Hou,Tao Li,Jihong Cao,Yijun Tan,Bu Deng,Hongyi Li,Jun Zhang,Yaping Gu
DOI: https://doi.org/10.1088/1742-6596/1828/1/012155
2021-02-01
Journal of Physics: Conference Series
Abstract:Abstract Here are many challenges for target detection in the complex scenes. In the real scene, such as the diversity of the type, quantity and shape of the parts on the workpiece plates, the model construction and rapid detection need to deal with the complex and changeable scene, and the large-scale massive point cloud data challenges the operational efficiency and accuracy of visibility analysis technology. In this paper, the 3D modeling method based on computer graphics is proposed by using the displacement laser sensor as the digital representation of 3D complex scene on the scanning point cloud, which realizes the fast extraction of semantic target feature parameters under complex conditions. The multi-dimensional parameter-based saliency feature and fusion multiple filtering algorithm is built to obtain a fast and high-precision intelligent template recognition detection method. Based on SOC embedded engineering technology, the detection algorithm is realized by software programming. The two-dimensional mechanical motion platform is developed to test the workpiece plate applied to wheel hub of high-speed rail train, and the recognition accuracy is more than 99%.