Research On Moving Object Tracking For UUV Recovery

Unmanned Underwater Vehicle (UUV) is getting more and more attention in various countries because of its great value, meanwhile, the underwater recovery of UUV technology research is increasingly concerned by people. Visual target tracking belongs to the high-level research in computer vision, and has become a hot topic in the field of vision. This paper adopts the piggyback recovery method, in close range recovery stage, UUV uses monocular camera for visual tracking the guided lights of submarine's back; Because of the complexity of underwater environment and conditions, such as noise and interference, it is important to study visual target tracking algorithm of UUV. This paper makes the research of visual moving target tracking method during UUV recovery process from the following aspects.Firstly,this paper introduces monocular visual tracking system for UUV undersea piggyback recovery. According to the actual situation, we choose UUV undersea directly piggyback recovery and introduce the hardware components. By using a linear light array, we can get UUV 4DOF positioning information. By analyzing the images taken at different azimuth, we can get the target light source position in the world coordinates.Secondly, the underwater image is pretreated and the target light source is extracted. For the presence of periodic noise, using the Gaussian band-stop filter for frequency domain filtering. To enhance underwater image in the spatial domain, we adopt piecewise linear transformation to improve image contrast, underwater image smoothing and sharpening processing. For image motion blur phenomena, this paper analyzes its degradation model and takes Wiener filtering for recovery processing. At the same time, the improved Otsu segmentation method is adopted, then we use the method based on Bayesian decision theory to remove pseudo light source.Thirdly, visual target tracking algorithm of UUV is studied and discussed. Considering the interference of target tracking, Mean Shift tracking algorithm based on the weighted target region and weighted background is proposed to improve the accuracy of target tracking.When the moving target scale changes, because the tracking window of Mean Shift tracking algorithm remains the same size in the whole process, this paper adopts a tracking algorithm based on adaptive scale; By getting the center of the heart-shaped target light source and adjusting the bandwidth of kernel function, so the tracking window can be adaptive updated when the scale of target changes.Finally,this paper is verified by simulation. We use 11-DOF platform and other equipments for semi-physical simulation, then design the visual tracking simulation experiments: the guided light array do rectilinear motion and sinusoidal motion. By analyzing the experimental results we can verify the effectiveness of these methods.