Design And Implementation Of Object Recognition System Based On SIFT Feature

With the growing of the modern warehouse and the variety of the goods, the demand in the process of warehouse management arises gradually, which is to identify the object by an image, and how to recognize objects is becoming more and more difficult for warehouse keeper. So this paper wants to solve that problem by using the object recognition technology based on computer vision.SIFT(Scale-invariant feature transform) feature is a very effective feature extraction method. This paper built an object recognition system based on the SIFT feature. This paper firstly analyzed how to extract SIFT and DSP-SIFT feature, and carried out a comparison experiment which found out that though DSP-SIFT feature’s matching is slightly better than SIFT, its time complexity is much higher than SIFT. Then, this paper analyzed the matching technology based on SIFT feature, and carried out a matching experiment which found out that SIFT has a good robustness to rotation, perspective changes, block and brightness changes.In the object recognition system based on SIFT,firstly, this paper proposed the bag of words model based on the combination of SIFT and Dense-SIFT feature according to their advantages and disadvantages in bag of words model. Secondly, for bag of words model ignores the information of object’s spatial structure, this paper proposed a spatial pyramid model based on the combination of SIFT and Dense-SIFT feature. Thirdly, RBF kernel function was used in this paper with SVM classifier, after its parameter’s optimization. Finally, a group of simulation experiment was performed to verify the classification performance of space pyramid model.Finally, by using the OpenCV and Qt open source software library, an object recognition system was designed and implemented in this paper.