Research And Implementation Of Mmwave Point Cloud Processing Algorithm For Personnal Perception

With the in-depth development of 5G,millimeter wave radar technology is gradually being used in various identification and perception scenarios.Among them,person perception is a kind of perception method that recognizes various features of people through their biometric characteristics.In application scenarios such as smart home,autonomous driving,security protection,etc.,millimeter wave radar people sensing technology has played an important role.Compared with traditional people sensing techniques,millimeter wave based people sensing approaches have the characteristics of high privacy and confidentiality,can be monitored for a long time,and are not affected by light and environment,thus compensating for the shortcomings of vision-based techniques.However,existing studies use point clouds that tend to be directly output by algorithms encapsulated inside commercial millimeter wave radars,which have problems such as low number of point clouds and limitations of quality fluctuations.To solve these problems,this paper optimizes the algorithms related to point cloud processing and proposes a person-aware application for home scenarios.The method is able to identify the person trajectory and further protect the home security.In this paper,the original dataset of person trajectories is collected and the signals applicable to person trajectory acquisition are modulated.Then,a fast fourier transfonn technique is used to realize the transformation from raw data to point clouds,and an optimized dynamic target detection algorithm is proposed,which has improved the quality of point clouds to a certain degree.Then,point cloud clustering and trajectory tracking algorithms are used to obtain continuous point cloud sequences and set rules for personnel trajectory screening.Finally,a person classification model combining spatial transformation network and convolutional neural network is designed,and triplet loss is introduced to achieve stranger perception in the open set case.This algorithm is deployed and tested on a commercial millimeter wave radar,and after experimental evaluation,the results show that the optimized point cloud data in this paper improves the target recognition accuracy by about 2.25%.At the same time,the proposed human perception model achieves a classification accuracy of 96.67%for known categories and 87%for stranger recognition accuracy on this test set.