前方车辆检测的特征融合算法研究与实现
收稿日期: 2016-02-06
网络出版日期: 2017-12-30
基金资助
国家自然科学基金资助项目(61376028); 上海市经济和信息化委员会资助项目(11XI-15)
Feature fusion for front vehicle detection and implementation
Received date: 2016-02-06
Online published: 2017-12-30
为了解决前方车辆检测的鲁棒性和实时性问题, 提出了一种基于车辆形态特征和类HAAR 特征融合的前方车辆检测优化算法. 为了克服车底阴影提取易受外部环境因素影响的缺陷, 采用猴王遗传算法(monkey king genetic algorithm, MKGA)进行阈值分割, 提取车底阴影部分; 然后通过车辆形态特征一次筛选得到感兴趣区域, 并对感兴趣区域的类HAAR 特征进行提取和降维, 输入支持向量机(support vector machine, SVM) 训练好的汽车分类器进行二次筛选. 随机抽取视频的300 帧进行算法验证, 实验结果表明: 算法在复杂环境下能够实现车辆检测, 并且相比于单一特征的检测方法, 准确率由80% 提高至90%; 利用类HAAR 特征积分图和主成分分析(principal component analysis, PCA)降维能够有效地提高检测速度. 算法满足驾驶辅助系统准确性和实时性的要求.
刘冬军, 徐美华, 龚露鸣, 夏臣君 . 前方车辆检测的特征融合算法研究与实现[J]. 上海大学学报(自然科学版), 2017 , 23(6) : 893 . DOI: 10.12066/j.issn.1007-2861.1767
To achieve robust and real-time detection of front vehicles, an algorithm is proposed based on morphological features and HAAR-like features of vehicles. To avoid interference of external environmental factors in the extraction of shadows underneath the vehicle, the monkey king genetic algorithm (MKGA) threshold segmentation is used. Regions of interest are obtained using morphological characteristics in the initial screening. HAAR-like features of these regions are extracted, and the dimension is reduced. In the secondary screening, an auto classifier trained by support vector machine (SVM) is input. A total of 300 frames are randomly selected to verify the algorithm. Experimental results show that the algorithm can detect vehicles in a complicated environment. Compared to single feature detection, accuracy is raised from 80% to 90%. The HAAR-like integral diagram and PCA for dimension reduction can effectively improve detection speed. The algorithm meets the driving assistance system requirements both in accuracy and speed.
[1] 郑湃, 崔立真, 王海洋. 云计算环境下面向数据密集型应用的数据布局策略与方法[J]. 计算机学报, 2010(8): 1472-1480.
[2] 胡铟, 杨靖宇. 基于模型的车辆检测与跟踪[J]. 中国图像图形学报, 2008, 13(3): 450-455.
[3] Sun Z H, Bebis G, Miller R. On-road vehicle detection: a review [J]. IEEE Transaction on Pattern Analysis and Machine Intelligence, 2006, 28(5): 694-711.
[4] Wang J X, Bebis G, Miller R. Overtaking vehicle detection using dynamic and quasi-static background modeling [C]// Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. 2005: 64-71.
[5] Franke U, Gavrila D, Grzigs, et al. Autonomous driving goes downtown [J]. IEEE Intelligent Systems & Their Applications, 1998, 13(6): 40-48.
[6] 慕永云, 王荣本, 赵一兵. 基于多特征融合的前方车辆检测方法研究[J]. 计算机应用研究, 2011, 28(9): 3572-3575.
[7] 张建明, 张玲增, 刘志强. 一种结合多特征的前方车辆检测与跟踪方法[J]. 计算机工程与应用, 2011, 47(5): 220-223.
[8] 金立生, 王岩, 刘景华, 等. 基于Adaboost 算法的日间前方车辆检测[J]. 吉林大学学报(工学版), 2014, 44(6): 1604-1608.
[9] Otsu N. A threshold selection method from gray-level histogram [J]. IEEE Transaction on Systems, Man and Cybernetics, 1979, 9(1): 62-66.
[10] Wang L, Duan H CH, Wang J L. A fast algorithm for three-dimensional Otsu’s thresholding method [J]. IEEE International Symposium on IT in Medicine and Education, 2008, 153(2): 136-140.
[11] 陈琪, 熊博莅, 陆军, 等. 改进的二维Otsu 图像分割方法及其快速实现[J]. 电子与信息学报, 2010, 32(5): 1100-1104.
[12] 汪海洋, 潘德炉, 夏德深. 二维Otsu 自适应阈值选取算法的快速实现[J]. 自动化学报, 2007, 33(9): 968-971.
[13] 王宏文, 梁彦彦, 王志华. 基于新遗传算法的Otsu 图像阈值分割方法[J]. 激光技术, 2014, 38(3), 364-367.
[14] 李宇中, 刘红星, 张胜. 猴王遗传算法的改进[J]. 南京师范大学学报(工程技术版), 2004, 4(3): 53-56.
[15] 姜兰池, 沈国强, 张国煊. 基于HSV 分块颜色直方图的图像检索算法[J]. 机电工程, 2009, 26(11): 54-56.
[16] Violas P, Jones M. Robust real-time objects detection [J]. International Journal of Computer Vision, 2004, 55(2): 137-154.
/
| 〈 |
|
〉 |
