Research On Key Technology Of Environment Perception During Autonomous Landing Of Mars Probe

Environmental awareness plays an important role in navigation,guidance and control system in the landing process of Mars probe.According to the sensor types,environmental perception technology can be classified into two categories: one is that the perception technology based on active sensor.These sensors transmit radio waves or light signals to the target area and receive the reflected signals for environment perception.This kind of sensors mainly include LIDAR,microwave radar and so on.The other is that the perception technology based on passive sensor.These sensors obtain the visible or infrared signals reflected by the environment passively.Such sensors mainly include optical cameras,infrared cameras and so on.Passive sensors have the advantages of light weight,small size,low power consumption and abundant sensing information,while active sensors are not restricted by illumination conditions and can obtain accurate three-dimensional information.Combining the advantages of active and passive sensors,considering the working range of sensors and the requirement of sensing tasks,this paper carries out the related research work of environment sensing based on active and passive sensors in autonomous landing process of Mars probe.First,the existing problems in navigation,guidance and control schemes are analyzed in autonomous landing process of Mars probe,and a scheme of applying environmental sensing technology to GNC system is proposed.The research task,research object,system output and research environment of environmental awareness are defined,and the architecture of environmental awareness system is constructed.According to the system research scheme,the key technologies of environmental perception research are refined.Second,in view of the impact of Mars dust weather on the environmental perception based on optical images during landing,a clear image restoration method is proposed in Mars dust environment.Firstly,the image degradation model is established for images captured in dust environment,and the problem of image clarity in Martian dust environment is transformed into the problem of calculating atmospheric light value and transmission coefficient value in the model.For the calculation of atmospheric light value,a method is designed based on quadtree subdivision.This method searches the specified threshold area with maximum average on the minimum value image by division iteration and finds the corresponding region in the input image.The mean values of each channel in the region were calculated as the atmospheric light value.For the transmission coefficient,the obtained atmospheric light value and dark channel prior were used to calculate the transmission coefficient.The experiments show that the proposed image restoration method has a certain adaptability for illumination change,dust intensity change and scene change.Compared with other restoration methods,the proposed method has obvious advantages in subjective evaluation and objective evaluation.Third,in the absence of global satellite navigation and positioning system support,a position sensing method is proposed based on matching between descent image and reference image.On the basis of feature points detection,a high-dimensional feature space index structure model and the corresponding matching query method were designed based on improved k-d tree,which avoid the inefficiency of matching by exhaustive method and solve the problem that k-d tree is not suitable for matching query over 20-dimensional space.On this basis,the matching points are precisely matched by the maximum-likelihood consistent estimation method,and the conversion relationship between the descent image and the reference image is calculated by the precise matching feature points,then the position of the detector is calculated on the reference image.Simulation results show that the proposed method can effectively determine the location of the detector.At the same time,the experiments were carried out using real images taken by Mars orbiter.The experimental results further verify the effectiveness of this method.Fourth,in order to avoid the detector falling into the impact craters during landing,a method of detecting and recognizing impact crater areas is proposed based on optical image.Aiming at the shape characteristics of impact craters,a method of edge detection is proposed based on structure learning.The edges of impact craters are learned and detected by using the extended form of random forest,i.e.structured random forest.The decision trees are trained in BSDS500 data set by means of transfer learning.For the detected edges information,a method based on edge groups similarity and a method based on morphology were designed to extract the candidate areas of impact crater.For the candidate areas of impact crater,the impact crater recognition methods are studied based on global Gist feature and deep learning model of AlexNet respectively.Experiments show that the method proposed in this paper can detect and identify impact crater areas effectively.Compared with other methods,the method proposed in this paper has relative high detection accuracy.Fifth,in order to avoid the collision between the detector and the larger rocks,a rock region detection method was studied based on saliency regions fusion of active and passive images.Firstly,the registration of active image and passive image was performed based on image region features and edge features.After registration,the saliency maps of optical image and depth map are calculated respectively.For the saliency map of optical image,it was calculated in color,brightness and direction channels and obtained by weighted combination.For saliency map calculation of depth map,the initial saliency map is calculated by the method of region segmentation.Then the energy function is constructed and optimized by max-flow min-cut method.The iteration is repeated to make the estimation of the saliency map more accurate.Two types of saliency maps are integrated into one new saliency map and the rock obstacle areas can be determined by the new saliency map.Finally,the recognition of rock area is identified by the same recognition method as craters.Experimental results show that this method can effectively detect the location of rock.Compared with other saliency detection methods that were frequently used,this method has better detection effect and higher accuracy than the other detection methods.Finally,a slope sensing method based on 3-D point cloud measurement data is proposed to avoid slide down or tip over caused by landing on a slope over tolerance for the detector.After coordinate transformation,a slope sensing method is designed based on sparse subspace clustering.The data point space is divided into slope subspace and plane subspace.The Euclidean distance from the point in subspace to the plane is calculated after establishing a spatial plane model.If the distance from a point to the plane exceeds the set threshold,this point is considered as an error clustering point and removed from the subspace.Correct clustering points are used for plane fitting,and the normal vectors are calculated according to the fitting planes.The angle between normal vectors is equal to the slope angle in geometric relationship.The experimental results of computer simulation,laboratory simulation and real Mars landform data show that slopes can be estimated accurately by this method.Compared with other methods,this method has relative small deviation.