基于卷积神经网络的道路车辆检测方法

发布时间：2018-01-20 23:51

本文关键词： 车辆工程车辆检测单目视觉卷积神经网络阴影分割　出处：《吉林大学学报(工学版)》2017年02期 　论文类型：期刊论文

【摘要】：提出了一种基于卷积神经网络的前方车辆检测方法。首先,根据车底阴影特征,运用基于边缘增强的路面检测算法以及车底阴影自适应分割算法来分割并形成车底候选区域,以解决路面灰度分布不均及光照条件变化问题;其次,运用针对道路交通环境的卷积神经网络结构,建立图像样本库进行网络训练;在此基础上,采用基于卷积神经网络识别的方法以验证并剔除被误检测为车底阴影的候选区域,进而确定真正的车辆目标;最后,修改网络为三分类识别,以验证本文方法的强扩展性的优势。实验结果表明:本文提出的车辆检测方法能够很好地区分车底阴影和非车底阴影干扰,有效地提高车辆检测的准确率和可靠性,降低误检率。
[Abstract]:A forward vehicle detection method based on convolution neural network is proposed. Firstly, according to the shadow features of the vehicle bottom. The road detection algorithm based on edge enhancement and the adaptive segmentation algorithm of vehicle bottom shadow are used to segment and form the vehicle bottom candidate region to solve the problem of uneven distribution of road gray and variation of illumination conditions. Secondly, using convolution neural network structure for road traffic environment, the image sample database is established for network training. On this basis, the method based on convolution neural network recognition is used to verify and eliminate the candidate area which is detected as shadow under the vehicle, and then determine the real vehicle target. Finally, the network is modified to three-class recognition to verify the advantages of the strong expansibility of the proposed method. Experimental results show that the vehicle detection method proposed in this paper can distinguish the shadow under the car from the shadow under the vehicle well. It can effectively improve the accuracy and reliability of vehicle detection and reduce the false detection rate.
【作者单位】：大连理工大学工业装备结构分析国家重点实验室;大连理工大学汽车工程学院;
【基金】：国家自然科学基金项目(61473057;61203171) 中央高校基本科研业务费专项项目(DUT15LK13)
【分类号】：TP391.41;TP183
【正文快照】： 0引言车辆间的剧烈碰撞是造成道路交通事故中人员伤亡的主要因素,前方车辆的实时准确检测有助于提前发现碰撞安全隐患,并及时向驾驶员发出警告,是实现防碰撞的关键技术之一。目前,单目视觉仍然是实现前方车辆检测的主要手段。主要原因在于其硬件的性能和价格合理,信息获取量大

【相似文献】