查询结果:   谭智勇,袁家政,刘宏哲,李青.基于深度卷积神经网络的人群密度估计方法[J].计算机应用与软件,2017,34(7):130 - 136.
中文标题
基于深度卷积神经网络的人群密度估计方法
发表栏目
人工智能与识别
摘要点击数
567
英文标题
CROWD DENSITY ESTIMATION METHOD BASED ON DEEP CONVOLUTION NEURAL NETWORKS
作 者
谭智勇 袁家政 刘宏哲 李青 Tan Zhiyong Yuan Jiazheng Liu Hongzhe Li Qing
作者单位
北京市信息服务工程重点实验室 北京 100101 北京成像技术高精尖创新中心 北京 100048    
英文单位
Beijing Key Laboratory of Information Service Engineering, Beijing 100101, China Beijing High-tech Innovation Centre of Imaging Technology, Beijing 100048, China    
关键词
人群密度估计 图像分块 深度卷积神经网络
Keywords
Crowd Density Estimation Image Block Deep convolution neural network
基金项目
国家自然科学基金项目(61271369,61502036,61571045);国家科技支撑项目(2014BAK08B,2015BAH55F03);北京市自然科学基金项目(4152018,4152016)
作者资料
谭智勇,硕士生,主研领域:数字图像处理,深度学习。 袁家政,教授。刘宏哲,教授。李青,讲师。 。
文章摘要
人群密度自动估计作为人群控制和管理的方法,是当前视频监控中的一个重要研究领域。现有的方法通过提取复杂的特征来进行人群密度估计,由于人群遮挡、透视效果和环境复杂等条件限制,难以满足实际应用中的需求,而深度卷积神经网络在特征学习上具有较强的能力。提出了一种基于深度卷积神经网络DCNN(Deep Convolution Neural Network)的方法来进行自然场景下人群密度估计。首先,为了消除摄像机透视效果,以图像中行人身高作为尺度基准,将图像分成多个子图像块。其次,设计一种新的深度卷积神经网络结构,利用多种不同的卷积核提取人群图像的深层次特征进行人群密度估计。实验结果证明该方法在自然场景下人群密度估计具有良好的稳定性和鲁棒性。
Abstract
Crowd density estimation is an important research topic in intelligent surveillance system, which is an effective way for crowd control and management. But the existing methods are hard to satisfy the demand of the practical applications, due to severe occlusions, scene perspective distortions and variable weather. In addition, most existing methods use general the hand-crafted features, which have low representation capability for crowd. To address these problems, a deep convolution neural networks (DCNN)-based method to estimate the crowd density in natural scenes is proposed. Firstly, we divide the crowed image into several image patches according to the criterion of the mean height of the adult pedestrian, which overcome the impact of perspective distortion on the pedestrian images Secondly, the deep convolution neural network has been designed. The DCNN is used to extract crowd features by different convolution kernels on the pedestrian image. The learned crowd features are employed to estimate crowd density. We test our approach on three different data sets, the experimental results demonstrate the effectiveness and robustness of the proposed method in the different scenes.
下载PDF全文