融合深度信息的BRISK改进算法

为了有效地利用RGB-D图像的深度信息,提高BRISK算法的尺度不变性和旋转不变性,提出一种融合深度信息的BRISK改进算法。首先,采用FAST算法提取特征点,并计算每个特征点的Harris角点响应值;然后,将整个图像划分为大小相同的网格,每个网格保留Harris角点响应值最大的特征点;其次,根据图像的深度信息直接计算特征点的尺度因子;最后,计算以特征点为中心的圆的灰度矩心,通过灰度矩心和特征点的位置偏差确定特征点主方向。从尺度不变性和旋转不变性两方面对几种算法进行了对比实验分析。实验结果表明,相比BRISK算法,改进后的算法在图像尺度变化时正确匹配特征点数提高了90%以上,在图像旋转时正确匹配特征点数提高了至少70%。

Improved binary robust invariant scalable keypoints algorithm fusing depth information

To effectively utilize the depth information from RGB-D (Red Green Blue and Depth) images and enhance the scale invariance and rotation invariance of BRISK (Binary Robust Invariant Scalable Keypoints) algorithm, an improved BRISK algorithm combined with depth information was proposed. Firstly, the keypoints were detected by the FAST (Features from Accelerated Segment Test) algorithm and their Harris corner response values were computed. Then, the entire image was divided into the same size grids, and the keypoint with the maximum Harris corner response value was reserved by each grid. Next, the scale factor of the keypoint was directly computed with the depth information of the image. Finally, the intensity centroid of the circle centered on the keypoint was calculated, and the orientation of keypoint was computed by the offset from its intensity centroid. The comparison experiment analysis of several algorithms on the scale invariance and rotation invariance was performed. The experimental results show that, compared with the BRISK algorithm, the number of correctly matched keypoints of the improved algorithm improves by more than 90% when the image's scale is changed and raises by at least 70% when the image is rotated.