| The development of the economy and road traffic technology has led to the rapid growth of super-large underground complexes,such as long-distance urban and intercity underground tunnels.Research on localization and mapping technology for trapped targets in closed space fires is of great significance for safety production and emergency rescue.Millimeter wave radar(MMW Radar)is an active microwave detection sensor that can detect targets through smoke and flames even in the absence of an external light source.This technology solves the problem of optical cameras and lidar losing their working ability in such environments.However,accurately identifying targets with single scan data can be challenging.By carrying the millimeter wave radar,a track robot can collect data from different positions in a fire scene.The simultaneous localization and mapping(SLAM)method can theoretically integrate data collected from multiple positions to construct a map that can identify targets,and address the issue of relying on external sensors for positioning in enclosed space fire scenes.Although various SLAM methods have proven to be capable of using two-dimensional(2D)millimeter wave radar to complete three degrees of freedom(Do F)pose estimation and mapping in open spaces,the complex multipath scattering phenomenon resulting from the closed space structure and target can cause the radar to generate a large number of false detections,leading to false targets,pose estimation errors,and even mapping failure.The specific research work is as follows:(1)A model was created to generate false detections from radar,and its impact on SLAM pose estimation was analyzed.A multipath scattering model for millimeter wave radar SLAM in closed space was established.By combining real closed space scenes and targets,the generation principles and differences of false detections were explained.The accumulation process and differences of single,double,triple,and multiple bounces in SLAM processing were also analyzed.The study further explored the influence of false detections on pose estimation and map construction.(2)Identification of false detections based on azimuth scattering angle features.Based on the differences in the scattering mechanisms between false detections and real point clouds,an azimuth scattering angle feature for radar was proposed.A method for calculating the azimuth scattering angle feature based on the equivalent scattering center of the target in the SLAM occupancy grid map was provided.The radar false detections can be detected based on the azimuth scattering angle feature.(3)Millimeter wave radar SLAM method combined with false detection identification.A SLAM method combining the Correlative Scan Matching(CSM)framework and false detection screening using the azimuth scattering angle feature was proposed.It solves the problem of distinguishing between multipath point clouds during the "scan to scan" process of the original CSM algorithm.A soft threshold STMR-CSM method was proposed to improve pose estimation accuracy by using reliable radar false detections in the "scan to submap " framework.Finally,the hard threshold HTMR-CSM method can eliminate false targets caused by false detections in the mapping process after obtaining high-precision pose estimation.Combining with the STMR-CSM method can improve both pose estimation accuracy and map construction accuracy simultaneously. |
