摘要
SIFT是特征提取与匹配技术中的一种有效的方法,具有较好的稳定性,以及旋转和尺度不变特性。但是SIFT特征提取与匹配的维数较高,且存在较大的误匹配率,影响双目立体视觉SLAM的实时性和准确率。为此,提出SMO-SIFT算法,对原SIFT进行欧氏距离比值的阈值选取进行粗匹配,再结合支持向量机的SMO算法,改进SIFT算法中的特征匹配算子。MATLAB仿真表明SMO-SIFT算法降低了算法的维数,改善了特征提取的实时性,同时提高了算法精确度,比较适合应用于双目立体视觉SLAM中。
SIFT is an effective method in feature extraction and feature matching technology, which has better stability and the features of rotation and scale invariant. However, SIFT represents high dimension of feature extraction and matching, and large mismatch rate, thus it affects the real-time and accuracy in binocular stereoscopic vision SLAM. Therefore, SMO-SIFT algorithm is proposed in this paper, which processed the coarse matching using threshold value of Euclidean distance ratio, and combined SMO of support vector machine into SIFT to improve feature matching operator. The MATLAB simulations prove that SMO-SIFT algorithm performs well on reducing dimension. It can impove real-time, increase accuracy simultaneously, and therefore is suitable to be applied in binocular stereoscopic vision SLAM.
引文
[1] Jiang Gang-yi, et al. Binocular vision based objective quality assessment method for stereoscopic images[J]. Multimedia Tools and Applications, 2014,74(18):8197-8218.
[2] Iatsun Iana, Larabi Mohamed-Chaker, Fernandez-Maloigne Christine. On the comparison of visual discomfort generated by S3D and 2D content based on eye-tracking features[C]. Proceedings of SPIE-The International Society for Optical Engineering, 2014:2136-2140.
[3] Wang Jian, Ni Yu-bo. Feature matching method for aircraft positioning on airdrome[J]. Lecture Notes in Computer Science, 2015,10(9218):284-291.
[4] M J Richard, C Danny. Live-Cell Tracking Using SIFT Features in DIC Microscopic Videos[J]. IEEE Transaction on Biomedical Engineering, 2010,59(7):2219-2228.
[5] 颜雪军,赵春霞,袁夏. 2DPCA-SIFT:一种有效的局部特征描述方法[J]. 自动化学报, 2014,40(4):675-682.
[6] Zergat, Kawthar Yasmine, Amrouche Abderrahmane. New scheme based on GMM-PCA-SVM modelling for automatic speaker recognition[J]. International Journal of Speech Technology, 2014,17(4):373-381.
[7] Gamboni Mo?ri, et al. Speeding up support vector machines: Probabilistic versus nearest neighbour methods for condensing training data[C]. Institute for Systems and Technologies of Information, 2014,364-371.
[8] Sinkar Sachin, Deshpande Ashwini. Object recognition with plain background by using ANN and SIFT based features[C]. IEEE International Conference on Information Processin, 2015:575-580.
[9] Liu Ning, Zhang Xiao-yu. An improved sift image matching detection[J]. Computer Modelling and New Technologies, 2014,18(10):282-287.
[10] Satnik Andrej, et al. A comparison of key-point descriptors for the stereo matching algorithm[C]. International Conference Radioelektronika, 2016:292-295.
[11] 许建华,张学工. 统计学习理论[M]. 北京:电子工业出版社, 2015.
[12] Redzuwan Redia, et al. Affine versus projective transformation for SIFT and RANSAC image matching methods[C]. IEEE International Conference on Signal and Image Processing Applications, 2015: 447-451.