Crowd Analysis And Airport Terminal Application Based On Deep Learning

The passenger traffic of China's civil aviation has been increasing in recent years,which has brought tremendous pressure to the management of airport terminals.The manual monitoring method of the traditional monitoring system is very time-consuming and laborintensive,and it is difficult to meet the demands of airport terminal safety management.Therefore,the research of intelligent monitoring technology is of great significance to the security of airport terminals and other public places.It is helpful for managers to adjust and dispatch security forces in a timely manner,allocate resources reasonably,and improve the ability to respond quickly to emergencies.The paper's research topic is crowd analysis based on deep learning and its application in the airport terminal.It focuses on two aspects: crowd counting and prediction of crowd density distribution.The widely used Euclidean loss in crowd counting assumes that the pixels are independent of each other and ignores the local correlation.Therefore,it is not sufficient for promoting the generation of high-quality density maps.To address these issues,we propose a crowd counting model based on the conditional generative adversarial network,which contains an adversarial loss to alleviate the blurring effect caused by Euclidean losses.Thanks to the game theory of generative adversarial network,we use Euclidean loss and adversarial loss to jointly optimize the model.It can strengthen the local correlation of the generated density image and enhance its quality,thereby improving the counting accuracy.Experimental results on two crowd counting datasets show that the model achieves comparable performance to mainstream methods.Most of the existing methods are not ideal for processing multi-scale information.In order to solve this problem,we propose a novel model for crowd counting and accurate density estimation,called enhanced scale robust network(ESRN).The embedded GAN module and the enhancer are designed to strengthen the local correlation of pixels and the robustness to scale variations.Extensive experiments on three crowd datasets show that the proposed model achieves competitive performance compared with the state-of-the-art methods.Finally,in view of the specific surveillance video data of the airport terminal,we propose a future crowd density distribution prediction model based on sequence modeling.The improved U-Net is utilized to model the sequence crowd density image,and then a discriminative network is employed to monitor it's output.The pixel intensity loss,adversarial loss,and optical flow loss are combined to optimize the predicted frame from the perspective of appearance and timing.Experimental results show that the model can effectively capture pixel motion information and achieve accurate prediction of crowd density frames at future moments while ensuring the image quality.