Research On Key Technology Of Machine Vision For Spacecraft Landing

Currently, it is an important aim of main astronautics countries to make spacecraft or probe safe landing on Lunar, Martian or other planetary surface in the field of space exploration. A key motivation for developing advanced Guidance, Navigation and Control system including environment perception, hazards detection, landing site search, path planning and autonomous motion control, is that communication latency and bandwidth limitations severely constrain the ability of humans to control robot functions remotely. Machine vision offers a good developmental direction for the establishment of autonomous landing Guidance Navigation and Control system of spacecraft. This paper researches machine vision-based algorithm applied in landing system of spacecraft, and focuses on three types of obstacles which are the security threat on the probe landing: craters, rocks and slopes detection algorithms.First of all, according to machine vision takes two major tasks - hazards detection and motion parameter estimation at the landing system of spacecraft, this part proposes a set of machine vision technical scheme. On this basis, the paper analyses work time of vision system and task goals. And then, workflow of machine vision-based algorithm is introduced based on the proposed landing Guidance Navigation and Control system. In according with spacecraft altitudes lowering and the sizes of different obstacles, hazards detection is divided into four stages: computation of smooth degree, crater detection, rock detection and slope estimation. This part also presents an algorithm of smooth degree which is based on Multi-scale Intensity Standard Deviation and experimental studies have shown that this method can be effectively completed primaries of the planned landing region.The second part of this paper proposes a crater detection algorithm for craeter-based predetermined landing site choice. This algorithm firstly detects crater candidate region in according with light-shade construct of crater image, and computes size of candidate regions. This step can reduce the crater search field greatly. And then, Bandeira's edge detection method and improved chord midpoint Hough Transform which reduces the computational redundancy are used to fit crater region by circle. Lastly, crater detection algorithm determines the extract region whether contains the crater by template matching, census transform, invariance and support vector machine. Virtual images validate the effectiveness of the algorithm.The third part of this paper gives the rock region detection for candidate landing site choice which is based on shadow detection. Maximum 2D entropy is used for shadow detection which can represent point-gray and small region-gray effectively and has a better anti-noise ability. On the basis of shadow detection, this paper demonstrates the method to obtain rock regions which is based on envelope model of the planet's surface rocks shape, light irradiation angle and the light incidence angle provided by the sun sensor. This method avoids ellipse fitting of shadow edge to reduce the complexity. Validation experiments show that the algorithm can effectively determine the rock areas.The fourth part of this paper studies feature tracking-based motion parameter estimation algorithm for landing path planning. On the basis of SIFT feature points extraction, RANSAC method is employed to estimate fundamental matrix F, and thus access to essential matrix E. Through the essential matrix E, the linear motion parameters estimated valuation is obtained and as the initial value of non-linear iteration. Simulation results show that the method has a certain degree of noise immunity. In this part, a method to obtain normal vector of plane is present firstly which employs the properties of homography induced by the planes and essential matrix E. On the basis of this method, this paper mainly presents a passive vision-based method of estimating the slope of the upcoming terrain wich is used for horizontal landing site choice and obtain final landing site. This method employs homography H induced by the planes, and SVD decomposition of essential matrix E which is obtained by fundamental matrix F and camera calibration matrix to compute two planes included angle. Under principle of homography matrix H compatible with fundamental matrix F, this method can obtain normal vectors of the planes in the first camera coordinate frame, angle of two planes, gradient of slope. Computer simulation and image experiments in the larborary demonstrate the accuracy and the validity of relative algorithms.Lastly, this paper gives an algorithm to select safe landing site which is based on hazard circles representing craters and rocks. Simulation results demonstrate the validity of the method. On the basis of slope estimation, this part also introduces a method to obtain horizontal landing site by computing largest connected plane. Compared with the method of horizontal plane choice by feature points corresponding, result of the method introduced in this part is better in simulation expriment. In according with the relative positions and the average speed of two frames, this part also researches a method of path planning which is based on acceleration and continuous variable thrust engine. Simulation results verify the feasibility of this approach.