Abstract: In order to detect the complex vehicle logo more availably, a new logo detection method for large vehicle using Edge Boxes is proposed. As a mature image segmentation algorithm, Edge Boxes can detect object proposals quickly and effectively. So, after the candidate regions of vehicle logo are located by the spatial relationship of the vehicle, Edge Boxes is applied to extract target regions feature. Then, a vehicle logo location classifier and a vehicle logo recognition classifier are constructed by Spatial Pyramid Matching based on Sparse Coding （ScSPM） and linear SVM. The target regions’ feature are trained and tested by two classifiers step by step to locate and recognize logo. Experimental datasets conducted by 4, 480 large vehicle images coming from 50 models which are captured by traffic surveillance system in various weather condition and illuminations. The experiments results demonstrate that the proposed method has great performance both in recognition accuracy and runtime, which shows the great potential for application.

Key words: large vehicle, vehicle logo detection and recognition, Edge Boxes, Spatial Pyramid Matching based on Sparse Coding（ScSPM）, vehicle logo location classifier, vehicle logo recognition classifier

摘要： 传统车标检测与识别算法难以检测大型车辆车标，且速度较慢。提出了一种基于Edge Boxes的大型车辆车标检测与识别方法。Edge Boxes算法是一种成熟的图像分割算法，能够快速且有效地检测物体位置，满足大型车辆车标检测与识别问题的准确性及实时性的需求。该方法首先根据车标在车辆中的空间位置关系初选车标候选区，然后利用Edge Boxes算法进行目标提取，进而将提取得到的目标送入利用线性约束编码构建的车标检测分类器和车标识别分类器进行训练与识别，得到车标检测与识别结果。对不同卡口的不同天气和光照条件下采集的4 480张图像（含50类大型车辆）进行实验，实验结果表明，在检测与识别性能以及时间消耗方面均优于传统方法，具有良好的实用前景。

关键词: 大型车辆, 车标检测与识别, Edge Boxes, 线性约束编码, 车标定位分类器, 车标识别分类器