交通场景中采用有监督序学习拥挤度排序算法

摘要

For automatic analysis of traffic scene attributes (‘congestion’,‘average speed’),traffic scene congestion de-gree ranking calculation model is proposed based on supervised learning.Using supervised learning ideas,a ranking function per attribute (‘congestion’,‘average speed’)is learned.For traffic congestion degree ranking model,Gist feature of each frame training images is extracted,however,for average speed degree ranking model,firstly,video motion information is ex-tracted and then Gist feature is extracted through frame differential method,finally,modified Ranking SVM projection model is introduced to learn a ranking function.This algorithm transforms traditional classification problem into training a ranking function on an attribute.Because this algorithm does not belong to hard division but belong to a finer measured model.Thus, this algorithm resolves the ambiguity consisted in traditional method of congestion degree estimation.Experimental results on three types of databases show that the proposed ranking model has relatively more accuracy and stability.%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。