基于LLC与加权SPM的车辆品牌型号识别

发布时间：2018-03-16 10:39

  本文选题：车辆品牌型号识别　切入点：方向梯度直方图　出处：《计算机工程》2017年05期 　论文类型：期刊论文

【摘要】：针对传统车辆识别算法鲁棒性及实时性不强的问题,结合局部线性约束编码(LLC)和加权空间金字塔匹配(SPM)模型,提出一种车辆品牌型号精细识别算法。提取图像方向梯度直方图特征,通过LLC对图像特征进行编码映射,得到具有语义信息的图像表达向量,以提高识别的准确率。利用加权SPM模型将空间位置信息引入图像表达向量中,并将每个图像的最终表达送入线性支持向量机分类器进行训练与识别。使用交通监控摄像头在不同天气和光照条件下采集150种车辆类型共56 827张图像进行实验,结果表明,该算法可有效改善识别效果,提高识别速度。
[Abstract]:Aiming at the problem that the traditional vehicle recognition algorithm is not robust and real-time, the local linear constraint coding (LLC) and the weighted space pyramid matching (SPM) model are combined. A fine recognition algorithm for vehicle brand model is proposed, which extracts the image direction gradient histogram feature, encodes and maps the image feature through LLC, and obtains the image expression vector with semantic information. In order to improve the accuracy of recognition, the weighted SPM model is used to introduce the spatial position information into the image expression vector. The final expression of each image is sent to the linear support vector machine classifier for training and recognition. The traffic surveillance camera is used to collect a total of 56,827 images of 150 vehicle types under different weather and light conditions. The results show that, The algorithm can effectively improve the recognition effect and speed.
【作者单位】： 中山大学工学院智能交通研究中心;广东省智能交通系统重点实验室;视频图像智能分析与应用技术公安部重点实验室;
【基金】：国家科技支撑计划项目(2014BAG01B04)
【分类号】：TP391.41
，

本文编号：1619562