基于DM642的嵌入式车牌识别系统设计与实现

摘要

In order to realize the real-time detection and recognition of Vehicle Plate in the embedded platform,this paper studies the Vehicle plate detection and recognition technology based on DM642.First,this paper uses a method which combines texture detection algorithm and Adaboost classifier to implement the vehicle plate location;then it uses maximum inter class distance method to correct the incline;segmenting vehicle plate characters by a method combining vertical projection and clustering connected domain;the recog nition process collects the Gabor filter feature of characters;this paper combine Genetic Algorithm and Support Vector Machine to opti mize Gabor parameters as the best parameters to distinguish different characters;and uses SVM to do the final classified training to get the final classification data.In addition,the algorithm of vehicle plate recognition system based on EMCV and Opencv image process ing library transplants codes in DM642.This system avoid the non-direction of Individual texture detection and the non-integrity of Individual Adaboost Classification in the recognition part.The recognition part can find the best Gabor parameters,which avoids the drawbacks of setting Gabor parameters without the maximum classification.After the identification,the data can be sent to the PC serv er based on the TCP protocol,implementing the location of vehicle plate and the identification of remote vehicles.%为了在嵌入式平台实现车牌的实时检测与识别,研究了基于DM642的车牌检测与识别技术.首先结合纹理检测算法和Adaboost分类器进行车牌定位,进而采用旋转差分最大类间距离法实现倾斜校正,校正后的车牌子图结合垂直投影法和聚类分析法对车牌字符进行分割,识别过程采集的是字符的Gabor滤波器特征,结合遗传算法和支持向量机优化Gabor参数作为区分不同字符的最优参数,并用SVM做最终的分类训练以得到最终的分类数据.车牌识别系统算法基于EMCV与Opener图像处理库在DM642上进行代码移植.该系统在车牌定位部分很好地避免了单独纹理检测定位的无方向性和单独Adaboost分类的定位不完整性.识别部分能找到最优的Gabor参数,避免了随机设置Gabor参数而无法最大分类的缺陷.识别后数据能基于TCP协议发送到PC服务器,实现远程车辆车牌的定位与识别.