Unmanned Aerial Vehicles (uavs) Based On Cpu, Gpu Sequence Image Registration Method Study Quickly

As a new remote sensing acquisition platform, UAV (Unmanned Aerial Vehicle) has received more and more attention and has been widely applied in many fields, due to its high resolution, high efficiency, low cost and low risk. In the processing of sequence image which acquired by UAV, it is necessary to ensure quality as while as timeliness. But Registration, as the first step also is essential step of UAV data processing, is always the bottleneck of processing.In this paper, from the Influence factors of the current UAV remote sensing platform’s imaging quality, we analysis the problems which are needed to be solved by UAV image registration through UAV image imaging features. By analyzing the problems which are needed to be solved by UAV image registration, we use color SIFT (Scale Invariant Feature Transform) as the ultimate method to resolve the problems. Color SIFT algorithm account of the invariance of color information on the basis of the original SIFT method, which accord with imaging features of UAV. Enhance the effect of image registration, reduce the feature confusion and mismatch that brought about by the traditional methods based on gray-scale image processing. Experiments show that number of the feature points which obtained by color SIFT algorithm more than seven times than the number obtained by the traditional SIFT algorithm. Color SIFT has more obvious advantages in more difficult to extract feature points in the image, such as Camera shot in the angled and large areas of water body in the Image. Through the advantages of GPU (Graphic Processing Unit) in graphics processing, the paper improves SIFT algorithm, efficiency of UAV data automatic registration and solves the problem of mass UAV data processing bottleneck. In CPU-GPU parallel computing system, computationally intensive steps will be processed in GPU. CPU-GPU system reduces the time of UAV image registration and enhances the efficiency of data processing. The increased efficiency from GPU is proportional to image size The bigger image is the more compute advantage gets from GPU. Processing speed based on CPU-GPU algorithm is61times quicker than CPU-based algorithm on dealing with images of3648x2736size. The advantage of the CPU-GPU computing combined with the effect of color SIFT algorithm enhance automatic registration efficiency of UAV data and provide an effective, feasible road for the rapid processing of UAV data.The main contents include:1. By analyzing the feature of UAV remote sensing platform and the imaging features of the UAV sequence image, Automatic registration for UAV CCD sequences image not only need to consider the geometric invariance, but also need to consider the color invariance. Color SIFT based UAV image registration algorithm addresses the geometric invariance of the image and also solve the problem of color invariance. Image strategy of block match is proposed and Experiments prove out this strategy can get higher match efficiency.2. Through the analysis of CPU-GPU computing platform and Design and implementation of the color SIFT algorithm, achieve Color invariance algorithm, geometric feature detection algorithm and the feature vector matching. Finally, analyze how to optimize the entire algorithm based on CUDA platform.3. Build hardware and software environment and design the workflow of experiments. Build a complete quickly processing of UAV sequence image From UAV data preprocessing to imaging with quadtree-based stripe block. Validate registration algorithm based on the color SIFT UAV image, color SIFT registration algorithm based on the CPU-GPU, UAV images quickly automatic processing, and analyze the result.