Keyframe-based real-time camera tracking

Abstract

We present a novel keyframe selection

Adaptive SIFT Matching Using Cascading Vocabulary Tree

“…From Fig. 7 we can see that in spite of how cascading vocabulary tree being constructed, our method all yields much more not only correct but also reliable matches compared with methods of [13,16]. Although vocabulary tree method of [16] achieves enough correct matching features, however, it cannot keep robustness when parameters are changed.…”

“…For demonstrating the correctness and the robustness of our method, we also compare our cascading keyframe selecting and feature matching with the method of [13,16] as illustrated in Figure 7.…”

“…All these images can be found on internet publicly. are image set from [16]. All these image sets has two parts: one represents reference images and the other presents live images.…”

“…By comparing with mean descriptors from top to down and at the same time scoring in each node, vocabulary tree based matching method can efficiently recognize objects in images due to the fact that visualization of vocabulary tree can be viewed geometrically as a nested voronoi graph, however, it is not an ideal tool to exactly match SIFT descriptors as k-d tree does. Dong et al [16] proposed an exact matching method that uses vocabulary tree to vote and then match features with selected keyframe, with which their method arrives at reliable matches even though the number of features increases. However, large amount of features in this method plays important role only in first voting stage, whereas fails to have any effect in second keyframe based matching stage, leading to the fact that it is only an image-image matching method in essence.…”

Object Detection Algorithm Based on Moving Scene

“…Du Dong Mei [9]proposed a fast SURF algorithm for save memory space. The algorithm transform integral image into one image space for saving memory space, Zhu Qiguang [10] proposed using wavelet method, the image is divided into high frequency and low frequency part, in order to improve the extraction performance of the feature points, Zhang Zilong [11] in the selection of the feature points on the proposed new the algorithm, to achieve a more accurate feature points matching.…”