3 12 6 Vol No 3 JOURNAL OF HARBIN UNIVERSITY OF SCIENCE AND TECHNOLOGY Jun 12 Harris SIFT 1 1 1 2 2 1 0001 2 0090 Harris SIFT SIFT 2 /3 Harris SIFT TP391 A 1007-2683 12 03-0069- 05 Harris Corner and SIFT Feature of Vehicle and Type Recognition KANG Wei-xin 1 CAO Yu-ting 1 SHENG Zhuo 1 LI Peng 2 JIANG Peng 2 1 College of Information and Communication Engineering Harbin Engineering University Harbin 0001 China 2 Heilongjiang Hashuang Road Management Office Harbin 0090 China Abstract Considering to the important factors of accuracy and real-time performance can not meet the requirement simultaneously in current vehicle recognition a compound image matching model and a recognition method were developed namely classifying vehicles firstly by utilizing Harris corner then classifying vehicles detailedly by using SIFT feature Compared with the one used SIFT feature only the method shortens the time of disposal by 2 / 3 under guaranteeing the accuracy to be kept essentially constant real-time performance was improved greatly Key words vehicle recognition Harris corner SIFT feature image matching 0 3 6 SIFT scale invariant feature transform 48 intelligence transport system ITS Harris 7 Moravec 1-3 4-5 Harris SIFT 8 11-11 - 25 1000005 1963 1985 E-mail yutin102@ 3 com
70 Harris SIFT Harris SIFT HSV hue saturation value Harris SIFT RGB red green blue HSV 11 1 if α I v x y B v x y β shadow x y = 1 &I s x y - B s x y ts 1 9 & I h x y - B h x y th 0 else 2 1 I h x y I s x y I v x y HSV H 混合高斯模型建立背景 视频获取 背景差分 移动区域获取 利用 HSV 空间检测阴影并去除 完成车辆提取 取无车一帧彩色图像为背景 S V B h x y B s x y B v x y HSV H S V ts th 11 10 α = 0 1 β = 0 63 2 a b c 3 待识别车辆 Harris 特征提取 样本车辆 Harris 特征提取 利用 Hausdorff 距离进行车型初步分类 待识别车辆 SIFT 特征提取 样本车辆 SIFT 特征提取 (a) 阴影检测背景图 利用特征匹配对车型再分类 识别结果 1 1 1 (b) 阴影检测原图 Stauffer 10 (c) 检测结果 2
3 Harris SIFT 71 2 3 3 8 1 1 2 Harris Harris 3 Harris Hausdorff Harris R 2 R = det C - ktr 2 C 2 SIFT SIFT C x = I 2 u x I uv x [ I uv x I 2 v x ] I u x I uv x I uv x I v x x u v k 0 4 0 6 Harris R T Harris 4 4 Harris 5 SIFT DOG difference 目标车辆小型车样本中型车样本 Harris 角点 Harris 角点 Harris 角点 of Gaussian DOG 建立角点图建立角点图 D x y σ = G x y kσ - G x y σ I x y 3 G x y kσ σ I x y L L x y kσ = G x y kσ I x y 4 根据最小值判定类型 5 Hausdorff m x y = L x +1 y - L x -1 y 2 2 槡 + L x y +1 - L x y -1 2 2 SIFT 5 SIFT θ x y = arctan ( L x y + 1 - L x y - 1 L x + 1 y - L x - 1 y ) 6 SIFT 8 1 SUV 2 1 Hausdorff Hausdorff 12 Hausdorff Harris Hausdorff 建立角点图 依次计算与 3 种样本的 Harsdorff 距离 样本库建立 大型车样本 Harris 角点 建立角点图 d 1 d 2
72 d 1 /d 2 Harris SIFT Harris Hausdorff SIFT SIFT 3 2 Harris 1 79 SUV SIFT 1 /3 2 2 6 待测目标 SIFT 特征 满足再分类条件? 初判定类别中样本 SIFT 特征 是 2 2 3 Core i3 2 2GHz CPU 2GB VS10 OpenCV2 2 3 Harris 56 8% SIFT SIFT 85 6% 0 81 9% Harris Harris 0 95 SIFT 5 24 1 73 /s 7 Harris 满足再分类条件? SIFT 否 7 a 小型车根据匹配点数 7 b 判定车型 6 SIFT 2 7 (a) (b)
3 Harris SIFT 73 SIFT 4% Harris SIFT 3 Harris 3 初分类车型 小型车 中型车 大型车 车型 细分类车型 轿车 SUV 小货车中型货柜车一般中型货车罐式车大型货柜车一般大型货车 3 样本总数 8 基于 Harris 角点方法的正确识别数 14 13 13 9 8 9 10 基于 SIFT 特征方法的正确识别数 19 19 本文方法初分类的正确识别数 57 37 58 本文方法细分类的正确识别数 Harris SIFT Harris SIFT - 1 SIFT 3 2 91-110 1 J on Computer Vision and Pattern Recognition Fort Collins 1999 04 21 1 78-80 2 J 08 24 11 288-290 3 J 07 10 2 62-65 4 LI Jian ZHAO Wang-zi GUO Hui Vehicle Type Recognition Based on Harris Corner Detector C / /Proceedings of the 2nd International Conference on Transportation Engineering ICTE Chengdu China 09 33-3325 5 Harris J 08 11 1 67-70 6 D 10 13-33 7 HARRIS C STEPHENS M A Combined Corner and Edge Detector C / /Proceedings of the Fourth Alvey Vision Conference Manchester the University of Sheffield Printing Unit 1988 147 8 LOWE D Distinctive Image Features from Scale-invariant Keypoints J International Journal of Computer Vision 04 60 9 XIA Li-min Vehicle Shape Recovery and Recognition Using Generic Models C / /Proceedings of the 4th World Congress on Intelligent Control and Automation Shanghai China 02 1055-1059 10 STAUFFER C GRIMSON W Adaptive Background Mixture Models for Real-time Tracking C / /Proceedings of IEEE Conference 2 246-252 11 CUCCHIARA R GRANA C PICCARDI M et al Improving Shadow Suppression in Moving Object Detection with HSV Color Information C / /Proceedings of IEEE Intelligent Transportation Systems Conference Oakland US 01 334-339 12 Hausdorff J 02 9 68-69