论文部分内容阅读
为了进一步降低虚假目标车辆的检测风险,提出了一种基于多元特征信息匹配的前方车辆图像识别方法。首先依据路面灰度均值突变搜索车辆候选区域,然后利用双通道Gabor滤波器提取车辆样本图像的多尺度方向特征,联合AdaBoost分类器与Cascade级联分类器形成一系列强分类器,对产生的5尺度8方向高维特征向量实施降维处理,同时分类筛选特征样本,最后结合灰度信息熵对称性测度辨识目标车辆存在性,完成了前方目标车辆的检测定位。研究结果表明:所提方法的检测准确率为96.7%,比经典算法提高了1.6%;整个检测过程最长耗时35 ms,最短耗时15ms,平均耗时25ms,检测耗时主要受车辆的大小以及背景复杂程度的影响;避免了单一特征下局部有效鉴别信息的损失,具有较好的识别精度和处理速度,车辆误检率仅为3.2%,优于其他车辆识别算法的误检率,提高了虚假目标检测的辨识度。
In order to further reduce the detection risk of false target vehicles, a method of vehicle image recognition based on multivariate feature information matching is proposed. Firstly, the vehicle candidate region is searched according to the sudden change of the mean value of the road surface grayscale, and then the multi-scale directional features of the vehicle sample image are extracted by using the two-channel Gabor filter. A series of strong classifiers are formed by combining the AdaBoost classifier and Cascade cascade classifier, Scale 8-direction high-dimensional feature vector is implemented, and at the same time, the feature samples are classified and screened. Finally, the existence of the target vehicle is identified according to the symmetry measure of gray-scale information entropy to complete the detection and localization of the target vehicle ahead. The results show that the detection accuracy of the proposed method is 96.7%, which is 1.6% higher than the classical algorithm. The maximum detection time is 35 ms, the shortest time is 15ms and the average time is 25ms. The detection time is mainly affected by vehicle Size and background complexity. It avoids the loss of locally valid identification information under a single feature, and has better recognition accuracy and processing speed. The false detection rate of the vehicle is only 3.2%, which is better than other vehicle recognition algorithms. Improve the false target detection identification.