基于贝叶斯推理的行人数量视频检测方法研究

发布时间：2018-08-31 21:06

【摘要】：本文针对行人数量检测问题进行了研究,在固定场景下,将行人图像分割为行人前景区,获取行人前景区的先验知识以及少量样本,使用贝叶斯规划方法推理出行人前景区中的行人数量,统计各前景区的行人数量得到图像行人数量。主要包括如下内容:1)对行人前景区分割过程中,首先使用背景模型进行过分割,为减少过分割带入行人前景区的影响,使用K临近区域快速合并方法进行了细致的分割,得到较为准确的行人前景区。2)对行人前景区进行特征提取,考虑了透视变形对特征提取的影响,采用基于固定目标的自标定方法生成透视变换矩阵,消除透视变形对特征提取的影响。计算行人前景区的熵特征和区域协方差特征,这些特征与行人前景区的尺寸无关,是基于前景区内容差异性的“基元”特征。3)在使用贝叶斯规划方法时,获取到行人前景区面积会随行人数量的增加而增大的先验知识,根据行人前景区的特征建立贝叶斯网对行人数量进行推理,使用少量行人前景区样本建立行人数量检测模型。在实验部分,根据贝叶斯规划方法推理出前景区的行人数量,使用少量的样本学习模型,在测试时得到了较高的准确率。本研究提出的行人数量检测模型,可以检测等待过街区域行人的数量,对行人过街交通信号的配时进行实时优化;可以用于实时获取在公交站点等待乘客数量,优化公共交通调度方案;可以检测大型公共场所出入口人群数量,为人群的疏散提供决策。
[Abstract]:In this paper, the pedestrian number detection problem is studied. In the fixed scene, the pedestrian image is divided into the pedestrian front area, and the prior knowledge and a few samples of the pedestrian front scenic spot are obtained. Bayesian planning method is used to infer the number of pedestrians in the former scenic spots, and the number of pedestrians in the former scenic spots is counted to get the number of pedestrians in the images. The main contents are as follows: 1) in the process of pedestrian front scenic spot segmentation, the background model is first used for over-segmentation. In order to reduce the influence of over-segmentation into the pedestrian front scenic spot, the fast merging method of K-adjacent area is used to carry out the detailed segmentation. The feature extraction of pedestrian foreground area is carried out, and the influence of perspective deformation on feature extraction is considered, and the perspective transformation matrix is generated by self-calibration method based on fixed target. The influence of perspective deformation on feature extraction is eliminated. The entropy features and the regional covariance features of the pedestrian front scenic spot are calculated. These features are independent of the size of the pedestrian front scenic spot, and are based on the "basis" feature .3 of the content difference of the former scenic spot. To obtain the prior knowledge that the area of pedestrian front scenic area will increase with the increase of pedestrian number, a Bayesian network is established according to the characteristics of pedestrian front scenic area to infer the number of pedestrians, and a few samples of front pedestrian scenic spot are used to establish a pedestrian number detection model. In the experiment part, according to the Bayesian planning method, the number of pedestrians in the former scenic spot is deduced, and a small number of sample learning models are used to obtain a higher accuracy in the test. The pedestrian quantity detection model proposed in this study can detect the number of pedestrians waiting for crossing the street, optimize the traffic signal timing of pedestrian crossing, and can be used to obtain the number of passengers waiting at the bus stop in real time. It can detect the number of people at the entrance and exit of large public places and provide decision for the evacuation of public transportation.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP18

【相似文献】