基于多任务全卷积网络的人流监测系统

摘要

由于尺度变化问题,对图片中的人流数量进行估计具有较大的挑战,随着深度学习的发展,现有一些基于多列或多网络的神经网络模型来提取尺度相关的特征,以提升密度估计的精度,但是,这些模型在进行优化训练时较为复杂,且需要消耗巨大的计算资源.鉴于此,本文提出了一种多任务全卷积网络来进行人流数量的估计,基于不同尺度的卷积操作,可以提取到尺度相关的特征,并同时对人流密度和人数进行估计,提高数据的使用效率,进而实现对高密度人流的估计.实验表明,所提的模型具有较好的精度和较高的鲁棒性.%Due to the change of scale,it is a challenge to estimate the number of people in the picture.With the development of deep learning,there are some neural network models based on multi-column or multi-network to extract the scale-dependent fea?tures to enhance the density estimation However,these models are more complex in optimization training and require a huge amount of computational resources.In view of this,we propose a multitasking fully convolutional network to estimate the number of people. Based on the convolution operation of different scales,we can extract the scale-related features and estimate the population density and population at the same time,Use efficiency,and then realize the estimation of high density of people.Experiments show that the proposed model has better accuracy and higher robustness.