E-mail Alert
RSS
-
摘要:
为实现复杂动态背景下快速、准确地检测运动目标，提出一种改进二进制鲁棒不变尺度特征(BRISK)算法的运动目标检测方法。首先对图像进行分块，利用图像熵对图像块进行筛选；然后针对特征匹配过程中存在大量误匹配的问题，采用k近邻算法与欧氏距离进行特征匹配；最后通过改进的顺序抽样一致性算法进行特征点提纯，进一步完成背景运动补偿，从而利用形态学处理分割运动目标。采用多组视频图像进行验证，本文算法在原BRISK算法的基础上去除了32.7%的特征点，并且匹配效率提高了75%，处理速度比以往算法快，并且具有较强的抗噪性能。
Abstract:In order to realize fast and accurate detection of moving targets under complex dynamic background, a moving object detection method based on BRISK (binary robust invariant scalable keypoints) algorithm is proposed. Firstly, the image is divided into blocks, and the image blocks are filtered by using image entropy. Then, aiming at the problem of large number of mismatch in the process of feature matching, the k-nearest neighbor algorithm and Euclidean distance are used to perform feature matching. Finally, the improved sequential sampling consistency algorithm is used to refine the feature points and further completes the background motion compensation, and morphological processing is used to segment the moving target. Through the verification of multiple video images, the proposed algorithm removes 32.7% of the feature points based on the original BRISK algorithm and improves the matching efficiency by 75%. The proposed algorithm has faster processing speed than previous algorithms and strong anti-noise performance.
-
Key words:
- moving object detection /
- BRISK algorithm /
- image entropy /
- Euclidean distance
-
Overview: Moving object detection has been the focus of research in the field of machine vision and intelligent transportation. Its purpose is to segment the moving objects from the sequence of video images so as to make the next step for target recognition, tracking and navigation. However, under the complex dynamic background, many factors such as light changes, background interference, camera motion and so on, make the detection very poor. At present, the commonly used feature point detection algorithms include SIFT, SURF and ORB algorithm, but they cannot meet the requirements of moving target detection. BRISK algorithm has better rotation invariance, scale invariance and better robustness. BRISK algorithm is the best one in the image registration with larger blur, but the real-time performance of BRISK algorithm is worse than ORB algorithm. Aiming at the real-time and accuracy of moving object detection algorithm, this paper proposes a moving object detection algorithm based on improved BRISK feature matching. Firstly, the video frames are divided into blocks, the entropy of each sub-block is calculated. The sub-blocks are filtered by using the image entropy, so that sub-blocks whose local information is too concentrated can be removed so as to avoid the influence of excessive local feature points. Secondly, the AGAST algorithm is used to detect the feature points of the remaining sub-blocks and generates the corresponding feature descriptors. Then, the feature matching is performed according to the k-nearest neighbor algorithm, and the feature point pairs are further purified by the Euclidean distance. So as to achieve the purpose of further improving the accuracy of the algorithm, and provide reliable data for calculating the next motion parameters. An improved PROSAC method is used to extract the optimal feature points to estimate the background motion parameters, and the background motion compensation is completed by combining the six-parameter affine model. Finally, the frame difference method and morphological process to extract the moving target, and the Otsu method is used to obtain the optimal threshold to achieve a more complete segmentation of the moving target. In order to evaluate the detection effect of the algorithm, three groups of video images are used to verify the algorithm. The proposed algorithm removes 32.7% of the feature points and improves the running time of 1.1 s based on the original BRISK algorithm. The detection efficiency is better than the previous ORB algorithm to some extent, at the same time improves the matching efficiency to more than 75%, and enhances the anti-noise performance of the algorithm. The experimental results show that the proposed algorithm can improve the real-time performance and ensure the robustness of the proposed algorithm. Compared with the previous detection algorithms, this algorithm is more suitable for the detection of moving objects in the complex dynamic context.
-
-
图 3 三组视频图像。(a)第一组视频序列V1；(b)第二组视频序列V2；(c)第三组视频序列V3
Figure 3. Video images. (a) The first set of video sequences V1; (b) The second set of video sequences V2; (3) The third set of video sequences V3
表 1 不同算法特征点数目及耗时对比结果
Table 1. Numbers of feature points and time-consuming of different algorithms
Video images Feature points/(time-consuming/s) SIFT SURF BRISK Proposed method Reference [7] V1 3639/18.96 1266/5.01 1217/2.60 584/1.06 295/1.06 V2 2378/14.52 986/4.08 610/2.423 374/0.98 328/0.94 V3 488/10.18 401/3.20 95/1.67 64/0.57 262/0.95 
下载: 导出CSV
表 2 特征匹配准确度对比结果
Table 2. Comparison of feature matching accuracy
Algorithms V1 V2 V3 Original featurepoints 584 374 64 Brute-force +RANSAC 253/56.7% 181/51.6% 32/50% KNNMatch 98/83.2% 90/75.9% 7/89.1% Proposed method 83/85.8% 58/84.5% 5/92.2%
下载: 导出CSV
-
[1] 吴克伟, 段伟伟, 杨学志.雨夜条件下的红外可见光视频融合目标跟踪[J].仪器仪表学报, 2016, 17(5): 1131-1139. http://d.old.wanfangdata.com.cn/Periodical/yqyb201605022
Yang K W, Duan W W, Yang X Z. Infra-visible video fusion object tracking under rainy night condition[J]. Chinese Journal of Scientific Instrument, 2016, 17(5): 1131-1139. http://d.old.wanfangdata.com.cn/Periodical/yqyb201605022
[2] Akshay S. Single moving object detection and tracking using Horn-Schunck optical flow method[J]. International Journal of Applied Engineering Research, 2015, 10(11): 30135-30152. http://cn.bing.com/academic/profile?id=01d9ecb9e07c3e17118cd3ec5f173eb1&encoded=0&v=paper_preview&mkt=zh-cn
[3] 刘洪彬, 常发亮.权重系数自适应光流法运动目标检测[J].光学精密工程, 2016, 24(2): 460-468. http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201602027
Liu H B, Chang F L. Moving object detection by optical flow method based on adaptive weight coefficient[J]. Optics and Precision Engineering, 2016, 24(2): 460-468. http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201602027
[4] 于红绯, 刘威, 袁淮, 等.基于子块运动补偿的运动目标检测[J].电子学报, 2017, 45(1): 173-180. doi: 10.3969/j.issn.0372-2112.2017.01.024
Yu H F, Liu W, Yuan H, et al. Moving object detection based on sub-block motion compensation[J]. Acta Electronica Sinica, 2017, 45(1): 173-180. doi: 10.3969/j.issn.0372-2112.2017.01.024
[5] 葛鹤银, 孙建红, 林楠, 等.融合小波变换及SIFT算法的去抖动运动目标检测[J].实验室研究与探索, 2016, 35(2): 119-123, 174. http://d.old.wanfangdata.com.cn/Periodical/sysyjyts201602030
Ge H Y, Sun J H, Lin N, et al. A moving target detection approach to remove jitter by fusing wavelet transform and SIFT algorithm[J]. Research and Exploration in Laboratory, 2016, 35(2): 119-123, 174. http://d.old.wanfangdata.com.cn/Periodical/sysyjyts201602030
[6] 唐佳林, 郑杰锋, 李熙莹, 等.航拍视频中运动目标检测算法研究[J].计算机科学, 2017, 44(S2): 175-177, 183. http://d.old.wanfangdata.com.cn/Periodical/jsjkx2017z2036
Tang J L, Zheng J F, Li X Y, et al. Research on detecting algorithm of moving target in aerial video[J]. Computer Science, 2017, 44(S2): 175-177, 183. http://d.old.wanfangdata.com.cn/Periodical/jsjkx2017z2036
[7] 刘威, 赵文杰, 李成, 等.基于改进ORB特征匹配的运动小目标检测[J].光电工程, 2015, 42(10): 13-20. doi: 10.3969/j.issn.1003-501X.2015.10.003
Liu W, Zhao W J, Li C, et al. Detecting small moving target based on the improved ORB feature matching[J]. Opto-Electronic Engineering, 2015, 42(10): 13-20. doi: 10.3969/j.issn.1003-501X.2015.10.003
[8] Leutenegger S, Chli M, Siegwart R Y. BRISK: binary robust invariant scalable keypoints[C]// Proceedings of 2011 IEEE International Conference on Computer Vision, 2011: 2548-2555.
[9] 何林阳, 刘晶红, 李刚, 等.改进BRISK特征的快速图像配准算法[J].红外与激光工程, 2014, 43(8): 2722-2727. http://d.old.wanfangdata.com.cn/Periodical/hwyjggc201408055
He L Y, Liu J H, Li G, et al. Fast image registration approach based on improved BRISK[J]. Infrared and Laser Engineering, 2014, 43(8): 2722-2727. http://d.old.wanfangdata.com.cn/Periodical/hwyjggc201408055
[10] Kriminger E, Cobb J T, Príncipe J C. Online active learning for automatic target recognition[J]. IEEE Journal of Oceanic Engineering, 2015, 40(3): 583-591. doi: 10.1109/JOE.2014.2340353
[11] 张闻宇, 李智, 王勇军.基于CenSurE-star特征的无人机景象匹配算法[J].仪器仪表学报, 2017, 38(2): 462-470. http://d.old.wanfangdata.com.cn/Periodical/yqyb201702025
Zhang W Y, Li Z, Wang Y J. UAV scene matching algorithm based on CenSurE-star feature[J]. Chinese Journal of Scientific Instrument, 2017, 38(2): 462-470. http://d.old.wanfangdata.com.cn/Periodical/yqyb201702025
[12] 崔少辉, 谢征, 王刚, 等.二进制鲁棒不变尺度特征匹配电子稳像[J].光学精密工程, 2015, 23(9): 2715-2723. http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201509035
Cui S H, Xie Z, Wang G, et al. Feature matching electronic image stabilization based on binary robust invariant scalable keypionts[J]. Optics and Precision Engineering, 2015, 23(9): 2715-2723. http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201509035
-
点击扫一扫
图(6)
表(2)
计量
- 文章访问数: 8843
- PDF下载数: 3311
- 施引文献: 0
