Study on the monitoring of pedestrian flow in transportation hub complex based on digital image processing
Student thesis: Doctoral Thesis
Related Research Unit(s)
With the socioeconomic development, the transportation develops rapidly in China. The high-speed railway, metro and other transportation modes have been popularized in more cities. The road network density of various transportation modes continues to increase. Therefore, the number of the Transportation Hub Complex is increasing significantly. The Transportation Hub Complex (THC) integrates the transportation, shopping, entertainment, leisure and other functions. Due to the complex internal structure, the huge flow of people and complex evacuation routes in the THC, it is very difficult to firefighting and rescue when fire is occurring. Moreover, the majority of such buildings is a local landmark. The fire and other emergency events will create an enormous negative social impact. In order to optimize the occupant load evacuation in fires and other emergencies, this thesis explored the method to dynamically monitor the occupant load in THC based on the computer vision technology. The framework of Occupant Load Monitoring System in Transportation Hub Complex based on image processing technology is proposed in this thesis. This system establish on network structure. The monitoring of occupant load in a region is realized by the processing and analyzing of multi-channel video image of video surveillance. In order to achieve the related functions of the system, a series of evacuation simulation experiments, video sampling experiments of pedestrian movement characteristics and field sampling in THC were conducted in this thesis. The videos of these experiments and field sampling are combined with the surveillance video recordings of the real scenes in the THC to establish the dedicated pedestrian image database for THC. The dedicated pedestrian image database provides the training set of the positive and negative samples. Based on the dedicated pedestrian image database, the algorithm of pedestrian recognition, detection and tracking which are suitable for THC were explored in this research based on the analysis of several typical algorithms of recognition and tracking of pedestrian targets. The recognition algorithm based on feature of Histogram of Oriented Gradients (HOG) in the Region of Interest (ROI) of upper and lower pedestrian body combined with the cascade Support Vector Machine (SVM) with four sub-classifiers based on the division of view angle were improved to achieve the detection of pedestrian in a THC in this thesis. For the pedestrian tracking problem, this study proposed a multi-feature template matching algorithm based on the combination of color luminance histogram feature and contour region feature. A template matching search strategy which is suitable for THC and a template update algorithm based on weighted sum method were improved for the template matching algorithm. The THC was divided into several regions and the occupant load in each region was monitored and summarized based on the algorithm and system proposed in this research. According to the dynamic statistical data of occupant load in each region, the dynamic statistical data of whole facility occupant load can be obtained. The dynamic statistical data of occupant load in each region and the whole facility is able to provide judging basis for the intelligent evacuation guidance system to organize optimization evacuation plan in the fires and other emergency situations. On the other hand, the algorithm and system proposed in this research can also be used to record the pedestrian movement behavior parameters in the key area. The scenario of the pedestrian passing through the multiple parallel bottlenecks is a typical pedestrian movement scenario in THC. Therefore, a series of pedestrian flow simulation experiments were conducted to investigate the pedestrian movement quantitative parameters of pedestrian passing through this scenario, such as time interval, flow rate and specific flow rate of pedestrian. Through data analysis of the velocity, time interval between two sequential pedestrians and flow rate, the effect of opening number, widths of opening and the space distance interval between adjacent openings in multiple parallel bottlenecks on those parameters are explored in this study. It is found that the pedestrian flow rate does not increase with the increasing number and total width of openings. It is related to the lane structure of pedestrians in the channel. And it is found that when the pedestrians pass through multiple parallel bottlenecks, there is an obvious right-hand opening preference. The discovery of these regularities, may give instructions for the design and improvement of the evacuation and pedestrian flow management facilities with multiple parallel bottlenecks, and will be useful for the development and verification of evacuation models. The study on the technology pedestrian monitoring in large-scale complex scenes based on video recording will be still in the development stage in a certain period. This study will provide a theoretical basis and technical support for the extensive application of the technology.
- Design and construction, Pedestrian facilities design, Digital techniques, Image processing, Transportation buildings