The Landing Technology Of Automatic Flying Object Based On Optical Flow

This master thesis is the project supported by Aeronautical Foundation Science ( No. 03153062 ).Visual motion analysis is a major field in computer vision, Its task is to make use of the motion image sequences on hand and pick up the features, including the points, lines or a set of lines, from the image, then according to the correspondence of the features, calculate the 2d motion character and estimate the 3d motion parameters or the structural information of the motional object. Visual motion analysis is broadly applied to industry and military, such as vision system of robot, resource prospecting, satellite tracking system and the firepower control system. This thesis applies the theory of visual motion analysis based on optical flow to automatic flying object landing process, track and recognize the target, estimate the motion parameters between the aircraft carrier and the automatic flying object, prepare for the landing of the flying object.Estimate 3d motion parameters and 3d structural information from image sequences consists of two steps. The first step is to pick up the image features. In this paper, we use the feature points as the image features. We can estimate the 2d motion information by the correspondence between the features. The second step is to establish a set of functions according to the rigid body motion function and the perspective projection, then get the motion parameters or the 3d structural information.In the first step, we introduce some common edge detection operators and compare the effect on our application. We also introduce the color image segmentation technology. The feature points matching algorithm we used is correlation matching strategy. It is easy to be carried out and make full use of the information of three color channels. In the second step, we use the 8-point algorithm to deal with the two usual motion modes which are the camera unmoving & the object moving or the camera moving & the object moving. We establish the mathematics model and compare the linear and non-linear algorithm. In the paper, all of our examples about the automatic flying object landing onto the aircraft carrier are made or simulated by 3DMAX5. The animations accord with standard of NTSC, It works like the television homing flying object. At the last, we discuss the errors including matching error, motion parameters error and depth error. We analyze the reason of the errors and give some useful suggestion on how to reduce the errors.