Abstract:

In the satellite robot ground assembly field, for the serious problem that the target identification algorithm based on scale invariant feature transform (SIFT) has real-time performance defect, a fast target -identification method is put forward combining with scalar quantization description and inverted file index (SQIF) to meet the need of accurate and efficient identification. On the basis of initial feature points detection, location and description, the scalar quantization of the floating-point description vector is completed through median division and covered mask. The query strategy of the inverted file structure is used to search neighbor feature points fastly. The stable matching points are selected based on the distance ratio criterion and random sampling principle. The center coordinate and boundary width of the rectangle box are solved according to the affine transformation result, and the typical fast target identification is completed. Experimental results show that, the mean average precision (MAP) curve surrounded area of the SQIF can reach 90.12% on average, at the same time, the SQIF can also accurate match feature points and box right target, which has the similar distinguish performance advantage to SIFT. However the execution time and overall time in the feature point matching phase account for 19.54% and 49.84% of that of the SIFT identification algorithm respectively, which has the real-time performance advantage.