SIFT算法在不同维数下的图像匹配效率

 在移动机器人视觉定位领域中，SIFT（Scale Invariant Feature Transform）算法由于具有尺度、旋转和光照不变等特性被得到广泛应用。图像匹配效率与SIFT算法中所取维数有关，使用多少维数来表示一个特征点，满足不同的图像匹配精度和实时性的要求，还没有明确定义。为解决这个问题，针对不同维数的SIFT算法的匹配效率做了实验分析，得到了不同应用场合下使用维数的指导范围。以室内场馆导航机器人视觉定位为例，选取了16维作为匹配维数，实验结果能够满足移动机器人实时定位的需要。

Efficiency of Sift Image Matching Algorithm of Different Dimensions

For mobile robot visual positioning, SIFT (Scale Invariant Feature Transform) algorithm is widely used because of its features of scale, rotation and illumination invariability. The efficiency of the SIFT image matching algorithm is related to the number of dimensions, which is taken as the characteristic point to meet different image matching precision and real-time requirements. To solve this problem, for different dimensions of the SIFT algorithm, experimental analysis was carried out for efficient matching to obtain a range of dimensions with a certain degree of practical significance. The test results show that, for image position, the characteristic point dimension functions are different. For a high performance of real-time image positioning, the function of 1×1×N is selected, and for the mobile robot image positioning, a high real-time but normal position precision is needed, the function of 2×2×N is a best choice. Taking the robot navigation of visual indoor positioning as an example, selecting 16 dimensions in a dimension matching of experimental results of the mobile robot, the needs of real-time location can be satisfied.