基于特征融合的车辆检测算法研究

发布时间：2019-02-16 01:47

【摘要】：近年来,随着交通事业的迅速发展,汽车数量也迅猛增加,道路交通安全问题日益突出,为了解决这一问题,社会各界致力于研究各种交通辅助系统,旨在通过智能交通系统来提高驾驶的安全性。车辆检测是智能交通系统中的重要组成部分,通过对驾驶环境的有效分析,对可能发生的交通事故进行预警,有效的降低交通事故的发生率。目前,基于特征的车辆检测方法被研究者广泛使用,这种方法主要分为特征提取和分类器训练两个部分。首先,在特征选取上要选择能有力表征车辆信息的特征,多层次垂直导向梯度和多层次局部二值模式都能较好的描述车辆的信息。其次,在分类器训练上要选用分类速度快,识别率高的分类器,交叉核SVM分类器具备这些优点。根据上述分析,本文设计了一种基于特征融合和交叉核SVM的车辆检测算法。本文设计了一种融合多层次垂直导向梯度特征和多层次局部二值模式算子的车辆特征,为避免融合后特征维数过高而造成训练模型时较为耗时,本文将主元分析应用到车辆检测系统中。用主元分析对多层次垂直导向梯度特征降维,然后再和多层次局部二值模式算子串联融合。本文使用交叉核SVM对得到的特征训练并分类,有效的缩短了训练生成模型和检测分类的时间。利用多尺度滑动窗遍历的方法实现对不同大小车辆的检测。这样就可能会导致同一目标车辆有多个检测结果,本文使用窗口融合的方法对多个检测结果进行融合,得到最终的检测结果。为验证本文算法的性能,本文做了对比实验,实验结果表明,本文的基于特征融合的车辆检测算法在有效的提升了车辆检测准确率的同时降低了虚警率,从整体上提高了车辆检测的性能。最后对本文的工作进行了总结,并对未来的研究工作进行了展望。
[Abstract]:In recent years, with the rapid development of traffic, the number of vehicles is also increasing rapidly, and the problem of road traffic safety is becoming increasingly prominent. In order to solve this problem, all walks of life are devoted to studying various traffic assistance systems. The aim is to improve the safety of driving through the intelligent transportation system. Vehicle detection is an important part of intelligent transportation system. Through the effective analysis of driving environment, early warning of possible traffic accidents can effectively reduce the incidence of traffic accidents. At present, the feature-based vehicle detection method is widely used by researchers. This method is mainly divided into two parts: feature extraction and classifier training. First of all, in the feature selection, we should choose the feature which can represent the vehicle information. The multi-level vertical guidance gradient and multi-level local binary mode can all describe the vehicle information well. Secondly, the fast classifier and the high recognition rate should be used in the training of classifier. The crossover kernel SVM classifier has these advantages. Based on the above analysis, this paper designs a vehicle detection algorithm based on feature fusion and cross kernel SVM. In this paper, we design a vehicle feature which combines multi-level vertical guidance gradient and multi-level local binary pattern operator. It is time-consuming to avoid the excessive dimension of feature after fusion. In this paper, principal component analysis is applied to vehicle detection system. Principal component analysis (PCA) is used to reduce the dimension of multi-level vertical guiding gradient and then merge with multi-level local binary pattern operator in series. In this paper, the cross-kernel SVM pair is used to train and classify the features, which effectively shortens the time of training generation model and detection and classification. Multi-scale sliding window traversal method is used to detect vehicles of different sizes. This may lead to multiple detection results of the same target vehicle. In this paper, the window fusion method is used to fuse the multiple detection results, and the final detection results are obtained. In order to verify the performance of this algorithm, the experiment results show that the vehicle detection algorithm based on feature fusion can effectively improve the accuracy of vehicle detection and reduce the false alarm rate. The performance of vehicle detection is improved as a whole. Finally, the work of this paper is summarized, and the future research work is prospected.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;U495

【相似文献】