CAJ | 학술논문

针对车牌无法识别的车辆，研究了一种车脸定位及识别方法。该方法分为两个阶段：首先，使用Adaboost算法进行车脸定位，并利用经验矩形方法进行定位改进；其次，在定位出来的车脸区域提取SIFT（scale-invariant feature transform）和SURF（speeded up robust feature）局部不变性特征，利用这两种不变性特征的叠加及位置约束改进匹配算法，与标准车型数据库中的车脸特征进行匹配，根据匹配结果进行车脸识别，从而得到车辆类型。实验结果表明，该方法的正确识别率达到83.6%。交通卡口抓拍到的车辆照片基本是正前照，无法获取车身侧面信息分析其车型。针对车牌无法识别的车辆，通过车脸定位、特征提取，并与标准车型库中车脸进行对比，进而识别车脸，该识别车脸的方法为识别车型提供了一种新途径。
침대차패무법식별적차량，연구료일충차검정위급식별방법。해방법분위량개계단：수선，사용Adaboost산법진행차검정위，병이용경험구형방법진행정위개진；기차，재정위출래적차검구역제취SIFT（scale-invariant feature transform）화SURF（speeded up robust feature）국부불변성특정，이용저량충불변성특정적첩가급위치약속개진필배산법，여표준차형수거고중적차검특정진행필배，근거필배결과진행차검식별，종이득도차량류형。실험결과표명，해방법적정학식별솔체도83.6%。교통잡구조박도적차량조편기본시정전조，무법획취차신측면신식분석기차형。침대차패무법식별적차량，통과차검정위、특정제취，병여표준차형고중차검진행대비，진이식별차검，해식별차검적방법위식별차형제공료일충신도경。
Focused on those vehicles with unrecognized license plate, this paper proposes a new approach for vehicle face localization and vehicle face recognition. The approach is divided into two steps: Firstly, adaptive boosting (Adaboost) algorithm improved by experience rectangle method proposed in this paper is used for vehicle face local-ization. Secondly, scale-invariant feature transform (SIFT) and speeded up robust features (SURF) local invariant features are extracted from the detected vehicle face, vehicle face recognition algorithm is improved by adding loca-tion constraints to matched points pair and combining SIFT and SURF. Then the vehicle type is recognized by com-paring it with the same features stored in the standard vehicle type database. Experiments show that the algorithm can locate vehicle face quickly and effectively, the correct recognition rate reaches 83.6%. Images captured on the highway bayonet are mostly from the front view, so the vehicle body side information to analyze vehicle type cannot be obtained. To those vehicles with unrecognized license plate, the approach by vehicle face localization, feature extraction and vehicle face recognition by comparison with each vehicle face stored in the standard vehicle type data-base provides a new solution for vehicle type recognition.