دانلود رایگان مقاله انگلیسی پارادایم محاسباتی تصمیم هوشمند برای نظارت بر جمعیت در شهرهای هوشمند - الزویر 2018

Abstract

The ever-expanding urbanization and the advent of smart cities need better crowd management and security surveillance systems. Advanced systems are required to improve and automate the crowd management system. The aim of the closed circuit television and visual monitoring systems using multiple cameras faces many challenges like illumination variance, occlusion and small spatial-temporal resolution, person in sleep, shadows, dynamic backgrounds, and noises. Therefore, the crowd monitoring, prevention of stampedes and crowd-related emergencies in the smart cities are major challenging problems. In this paper, we propose an intelligent decision computing based paradigm for crowd monitoring in the smart city. In the intelligent computing based framework, the optimization algorithm is applied to compute the feature of crowd motion and measure the correlation between agents based motion model and the crowd data using extended Kalman filtering approach and KL-divergence technique. The proposed framework measures the correlation measure based on extracted novel distinctive feature, and holistic feature of crowd data represent and to classify the crowd motion of individual. Our experimental results demonstrate that the proposed approach yields 96.20% average precision in classifying real-world highly dense crowd scenes.

Conclusion and Future Directions

In this paper, we have proposed a crowd management system for monitoring the crowd of the smart city. Agent motion-based learning model is applied for discriminatory features extraction of crowd motions of every individual from the crowd trajectories. An extended Kalman-filter based technique [36] and a KL-divergence based approach [35] [46] are used to evaluate the discrimination of the individual feature and the holistic features of crowd scenes. The extended Kalman-filter based approach mitigates the noises and provides the smoothed feature sets. We achieve the precision rate of 96.20% for the monitoring of crowd. Own experimental results show that the proposed crowd management system performs well in recognizing the individuals from the real world crowd scene. The shortcomings of this work are discussed below. We plan to address these issues in our future work.

• The proposed crowd management system is limited for the massive trajectory of several crowd motion inputs. It is a major issue to correlate between any crowd videos automatically. Manual estimations are required for perspective transformation, and in this case, we cannot entirely rely on multi-target tracking.

• The proposed system can be speeded if we use parallel programming paradigm. Even it can be implemented for some real-time application like visual camera surveillance. Finally, we plan to apply the deep learning based framework for crowd management and crowd tracking.