Object Detection And Localization Based On Fusing Multi-source Information And Path Planning Of Space Manipulator

The technology of object detection, tracking and locating based on photoelectric videos under the background of space environment is the key to realize the capture of hovering flight or support of docking for space station, etc. Object detection and localization based on fusing multi-source information and path planning of space manipulator mainly study that moving object detection and tracking based on two-dimensional image information from color camera, pose measurement of moving object based on three-dimensional space information from depth camera, and path planning of space manipulator based on the results of object detection, tracking and localization.According to the different backgrounds of space environment, this paper puts forward an index for evaluating image background based on information entropy, edge and color space to realize conversion of three modes, including land background, sea background and outer space background. In view of the complex background, such as land, this paper realizes keypoints extraction based on random selecting and object detection algorithm based on forward-backward optical flow. In view of the smooth background, such as sea background and outer space background, this paper realizes object detection algorithm based on the search strategy named "security domain" and the extracting method using background complexity and edge characters.To the detected object, this paper adopts the tracking method to achieve its extraction in image space. When the space station is in a state of stable-distance tracking, object tracking based on perceptual hashing with fragment-and-mean is realized, which can effectively solve the problem of global illumination and local illumination change. When the space station is close to the object with the state of capturing, deformable object tracking is achieved based on fusion of matching and optical flow, which can effectively solve the problem of object scale and rotation angle changes.To the tracked object, this paper realizes its three-dimensional pose measurement relative to the base coordinate of space manipulator based on matching algorithm with nonlinear scale space. Firstly through eye calibration, the transformation from color space coordinate to the depth space coordinate is realized; Then based on the eye-in-hand calibration, the transformation from deep space coordinate to the end coordinate of the space manipulator is realized; At last the three-dimensional location and pose measurement of the object relative to the base coordinate of space manipulator is measured. This location and pose is actually the one of the end of space manipulator, so we can use the kinematics inverse solution to calculate the amount of rotation of each joint when the space manipulator stretches, achieving manipulator path planning.