Lunar Obstacle Detection In The Process Of Lunar Lander's Landing

Lunar detection object of China consists of three steps: Orbiting, Landing and Returning. The Landing process has three steps. The first is to launch Lunar Lander. The second is to make it landing on the lunar surface. The last is to dispatch a lunar vehicle from the Lunar Lander to explore on the lunar. In this process, Landing on the moon, close to a predetermined target landing spot, in an area of rough terrain, is a difficult task. There are a lot of blocks, craters and steep slopes on the lunar surface, all these are called Obstacles. Accurate navigation relative to the lunar surface is necessary, together with the detection of possible hazards like these blocks, craters or steep slopes, Lunar Lander must keep away from these obstacles to search an area which is wide and flat to land. This paper makes a research to detect the environment of the lunar using image process and computer vision, we also make a simulation as a reference for the process of landing of Lunar Lander.The problem is that only relatively poor information about the detailed surface morphology of lunar is available. The lunar has been extensively mapped to a resolution of some 200m by NASA's Clementine mission. This is inadequate for a vision based Lander which must detect and avoid blocks and craters as small as the area of Lunar Lander. Because of this limit of the condition, the paper apples some perspective images to implement the simulation.Obstacles detection has three steps: The first step is separating and extracting the obstacle regions of the image. This paper adopts the method watershed which is based on mathematical morphology and gradient information. According to the different value of the object and background pixels, the method watershed uses the slight change of the edge value to get the goal of segmentation. Vincent etc firstly introduced the geography definition Watershed to image segmentation. They treated the gradient image as a supposing terrain surface and single pixel (or pixel gradient) as the altitude. For the edge pixels of the segmenting image often had large gradient value, which are called Watershed Lines and the minimize value of gradient pixels as the Valley of the surface. By detecting, he could gain the Watershed Lines, and this is the edge of the image. However, owing to the effect ofnoise and quantification error, the region inside emerge abundant local Valley. Every Valley of the gradient image will make a Catchments Basin during the immersion process, and this case is called Over-segmentation. Namely, a uniformity area can be divided into numbers of areas, and this process can produce so many false edges that we couldn't recognize the real edges. In addition, the flatten background will appear many Over-segmentation areas. The traditional method is to wipe off the noises before producing the gradient image, or choose a better method on gradient operator. The paper uses a watershed method which is based on multi-scale open and close operator. First, apply open and close operation once to the image. Second, put up the morphological gradient operation. Third, make a multi-scale mathematical morphology operation to the gradient image. Thus, we can weaken the effect which is produced by border edge during the operation. Minute border information could be preserved, and it also settles the problem that some false edges might appear at the big slope edge. Experimental results of the presented method indicate that it is efficient for lunar image segmentation.The second step is feature extraction. Carry on the characteristic choice and extraction to the target area towards the segmentation image of to carry on to remove to the false edge, outline and small district after removing etc. the processing take" pit" as an example, expected before in the picture processing work, extracted the edge of the target" pit", had the precise orientation to the position of" pit", so, position of" pit" has following several characteristics:(1) Area of" pit" The area of" pit" is bigger than one grey level, it is" the obstacle pit", accumulating the small ones in this level, is" the non- obstacle pit".(2) Perimeter of" pit" resembles paragraph (1)(3) Depth of "pit" The depth of" pit" can replace with a grey level (depth value).(4) Circularity The method contains the ratio of area and perimeter and the Hough transformation methods etc.At here, we can make use of the outline of the obstacle and its reflection to the initial image to get some feature descriptions, include the depth and the area of the obstacle, and get planar feature spaces.The third step is classification decision. Because the feature quantity that we choose is not too much, and the feature spaces are planar, it is not very high request that the choice of classifiers, underneath introduction the design step of the classifiers:(1) Build up the training set, we have already known that the point of training simples belong to which category. The source of The training simples is the image of the ground that is alike the condition of the lunar, and we need to carry on certain comparison's enlarge to the parameter in the process of feature extraction.(2) Set out from these conditions, look for a certain criterion function and linear discriminant function, and design to linear discriminant function.(3) Establish the model According to the parameter of the simples within the training set. The paper carries on the discrimination to more than tens training set that has already known the category, getting the maximal weight value.(4) Use this model for discrimination, make use of the criterion function and discriminant function to discriminate each position should belong to which category.Lunar exploration project is the hot topic of Chinese space technology. The paper put forward a obstacle detection system. Detecting the obstacles of the lunar and giving the technical reference for obstacle detection during soft landing of the Lunar Lander.