Intelligent Perception Algorithm For Small Body Environment Based On Multi-source Data Fusion

Accurate environmental awareness of unknown small bodies is a prerequisite for a safe soft landing of the probe.With the rapid development of artificial intelligence and multi-source data fusion technology,intelligent small body detectors will have more powerful environmental perception and adaptability.In order to overcome the problems of limited perceptual field of view,poor scene adaptation and anti-illumination capability of the current single observation means,this paper conducts an in-depth study on the intelligent small object environment perception algorithm with multi-source data fusion.The main research content of the paper is as follows.Firstly,an optical image-based algorithm for the intelligent detection of topographic features on the surface of small bodies is investigated,using a local variance equalization algorithm to enhance topographic features in low-illumination small body images,and data enhancement methods such as perspective changes are used to construct a benchmark dataset of topographic features on the surface of small bodies.For the dataset annotation error and missing problem,Using YOLOv4 combined with the self-learning network,the problem of labeling errors and missing labels in the dataset is effectively solved..At the same time,an adaptive cut-off network is designed to solve the small terrain feature miss detection problem.The experimental results show that the algorithm in this chapter improves the detection rate by more than 10% compared with the YOLOv4 network,and the comprehensive detection rate reaches 87.65%.Secondly,a 3D point cloud-based algorithm for intelligent detection of small bodies surface topography features is investigated.A simulated 3D model of the small bodies surface terrain is generated through the LIRS world building tool;the camera and LIDAR sensor are simulated and designed using Gazebo simulation software for image and 3D point cloud data acquisition.The simulation results show that the constructed3 D point cloud dataset can be effectively used in the experiments and that the PV-RCNN network used has a good detection rate in the absence of light conditions.Finally,in order to solve the problems of limited perceptual field of view,poor scene adaptation and anti-illumination capability of single observation means,an intelligent algorithm for small bodies environment perception based on camera and LIDAR with multi-source data fusion is proposed.The PV-RCNN network and the improved YOLOv4 network are selected as the basic network for target detection,and the transformation relationship between the lidar coordinate system and the image pixel coordinate system is used to project the 3D detection frame onto the image,and the weighted fusion algorithm is used to realize the 3D detection frame and the image.Optimal fusion of two-dimensional detection frames,and output reliable fusion results.The results of the comparison with the small bodies environment perception method of a single sensor show that the algorithm effectively improves the environmental perception ability of the small bodies probe and can adapt to the complex and diverse working environment.