Ground Target Localization Based On UAV Image And Remote Sensing Image

Targeting is a important application of UAV, it has become a hot research topic in one of the field of network processing and pattern recognition. With the further development of modern science and technology, targeting methods continued to emerge, such as radar law, the GPS method, But the positioning effect of these methods are device dependent, vulnerable to environmental impacts.Therefore, the paper proposes a ground targeting matching algorithm based UAV aerial images and remote sensing image matching algorithm. The algorithm aims to the nature of the aerial images and remote sensing images, extracting the common feature of the two images to matching the positioning. The experiments show that the algorithm for high positioning accuracy has a certain robustness to the image noise, light effects. Content and results of this study are as follows:The paper presents a ground-targeting framework using the combination of coarse matching and fine matching. First, using the block matching to find the target, re-use point features method for an exact match. UAV aerial images and satellite remote sensing images are coming from different sensors, the difference between the images will be very large. In this case, if using the point feature matching directly, the target location error will be larger, and there will be more mismatching points. Therefore, proposing the rough location of the block matching to find a target. The experiments show that the first use of block-matching method can increase the final positioning accuracy.The rough location of ground targets is based on the proposed algorithm of gradient vector flow (GVF) gradient amplitude histogram of the Mean Shift algorithm. Because of the color characteristics to the reference image and measured image prone to mutation,the traditional Mean Shift algorithm experimental results is not satisfactory. At the same time,when the Mean Shift algorithm is based on the orientation histogram or gradient direction histogram, there are some disadvantages in the two algorithms in varying degrees, for example, the matching surface is not smooth enough, the drift process is not continuous, the image histogram did not change after the object position has changed. To solve this problem, the paper presents the Mean Shift algorithm based on gradient vector flow (GVF) gradient direction histogram. The algorithm can effectively solve the above problems, and it is very robustness to image noise, illumination from the image.Using Surf algorithm to fine position the points of interest matching. The algorithm is robustness to image rotation, translation, scaling and noise impact. At the same time,the matching algorithm is better than sift. Then, utilizing RANSAC to calculate the coordinate transformation of the two images to obtain the transformation parameters. Experimental results show that simulation results verify the positioning accuracy of the results.Finally, in this paper, the design-based algorithm achieves the positioning of ground targets. Achieving the location of the target in remote sensing image. Through calculating the experimental error that is the mean and standard deviation of the results in each group, it can known that errors is within the acceptable range, the results of the positioning is very accuracy.