交通场景中采用有监督序学习拥挤度排序算法

发布时间：2018-01-17 03:17

本文关键词：交通场景中采用有监督序学习拥挤度排序算法　出处：《信号处理》2014年12期 　论文类型：期刊论文

【摘要】：为自动分析交通场景拥挤度与速度属性,提出基于有监督序学习交通场景拥挤度排序计算模型,利用监督学习思路分别学习交通拥挤度和平均速度两个属性的排序函数。在交通拥挤度排序模型中,首先提取每帧训练图像Gist特征,而对于平均速度排序模型,首先通过帧间差分法提取视频运动信息,然后再提取Gist特征,最后引入改进的Ranking SVM投影模型,学习得到每个属性排序函数。该算法把传统分类问题转化为关于某个属性训练一个排序函数,因不属于硬划分属于比较精细度量模型,从而解决传统拥挤度估计算法存在模糊性的问题。在三组交通视频数据集的实验结果表明本文的排序模型准确度、稳定性相对更高。
[Abstract]:In order to automatically analyze traffic scene congestion and speed attributes, a traffic scene congestion ranking model based on supervised order learning is proposed. In the traffic congestion sorting model, the Gist features of each frame of training images are firstly extracted by using supervised learning ideas to study the sorting functions of traffic congestion and average speed. For the average speed ranking model, the video motion information is extracted by inter-frame differential method, then the Gist feature is extracted. Finally, an improved Ranking SVM projection model is introduced. The algorithm transforms the traditional classification problem into training a sort function about a certain attribute, because it does not belong to the hard partition belongs to the more precise metric model. In order to solve the fuzzy problem of the traditional congestion estimation algorithm, the experimental results of three groups of traffic video data sets show that the ranking model in this paper is more accurate and more stable.
【作者单位】：燕山大学信息科学与工程学院;
【基金】：国家自然科学基金(61071199)
【分类号】：U491
【正文快照】： 1引言实时道路拥挤度信息将为交通出行计划和调整路线提供及时、可靠依据,对于缓解交通拥挤现象具有重要意义。近来,视频监控技术被广泛应用到交通信息管理中,监控网络规模越来越大,信息量越来越多,尤其在人口密集型城市更是如此,人工监控因为效率低难以实现,那么使用智能视频

【参考文献】