| As an important part of urban transport system,pedestrian traffic is a hot spot of present research in traffic study.However,pedestrian traffic is more individual and more diverse than vehicle traffic.In addition,pedestrian traffic is vulnerable to a variety of external factors and many features of that have a strong temporal-spatial domain.The extraction of pedestrian information was restricted to manual investigation or the video of cross-section processed by manual due to lack of detection means.It was unable to get temporal-spatial information and seriously affected the further research of microscopic and macroscopic mechanism for pedestrian crossing.These problems make it more difficult to describe the behavior of pedestrian crossing,and the corresponding mechanism modeling is backward and imperfect.In order to explain and model the mechanism of pedestrian crossing accurately,there is a need for extracting much more accurate and comprehensive temporal-spatial information.Therefore,this paper firstly adopts video image processing technology to realize the individual and group detection,identification,tracking which can provide information of the full temporal-spatial in the urban traffic environment.Secondly,a variety of macroscopic and microscopic characteristics of pedestrian traffic are analyzed on the corresponding external influencing factors and internal interaction.Based on the analysis of microscopic and macroscopic characteristics,the planning and execution mechanism model of the corresponding pedestrian crossing is constructed with reference to the latest research results of physiology and psychology.Finally,the model is validated combined with the actual traffic environment parameters.The results demonstrate that the pedestrian crossing mechanism based on the full temporal-spatial information can express the pedestrian crossing behavior more accurately,and thus providing the theoretical basis for the optimization with the temporal-spatial resources of pedestrian traffic.Specifically,the main contributions of this paper are listed as follows:(1)Optimizing target detection for pedestrian individual and group in urban traffic environment.In this paper,according to the characteristics of pedestrian detection and monocular camera in urban road monitoring,we firstly propose the C-HOG operator based on calibration constraint,which introduces pixel-pixel angle mapping detection area on basis of traditional HOG feature extraction.The algorithm limits the scaling range of the HOG matching and reduces the number of times of scaling.We establish the pedestrian template library,which realizes the improvement of the speed for the pedestrian’s individual detection without affecting the precision and improves the training library’s specificity.Secondly,based on the area projection features of camera calibration and combined with HOG detection,we adopt rectangular projection to segment pedestrian group so that we can complete detection of the pedestrian group by the optimal regression method.Thirdly,to resolve the problem of multitarget pedestrian tracking and occlusion in urban traffic,the method of multi-modal feature set combined with filter tracking and prediction is proposed.On the basis of pedestrian detection and recognition,the detected pedestrian target is input into the retrieval information base,and the corresponding feature subset is extracted by improving the SURF operator,dynamic edge and texture analysis.The method of fuzzy normalization is used to establish the multi-modal feature set of the information.Then,multi-target tracking and occlusion processing are realized by ?-? filter prediction and multi-modal feature matching.Finally,combined with consideration of the scene reconstruction algorithm and center offset,it is effective to complete the accurate extraction of multi-objective spatiotemporal trajectories in complex urban roads,which lays a solid foundation for the analysis of macroscopic and microscopic features.(2)Analyzing the macro-behavior feature of pedestrian crossing based on the full temporal-spatial informationBased on the extraction of full temporal-spatial parameters,the macroscopic parameters such as flow rate,velocity and density are obtained by means of data mining and pattern identification.Firstly,the spatiotemporal distribution of the pedestrian flow and the characteristics of the distribution are analyzed in depth,and the causes of the characteristics are also studied,and thus the corresponding arrival distribution model is established.Secondly,according to the analysis of the speed characteristics for pedestrians and the influence rules for gender,age,occupation and travel destination on the temporal and spatial distribution,we propose the corresponding feature expression models.Thirdly,the macroscopic characteristics of density and space occupancy for pedestrian crossing are studied,and then the causes of density characteristics are analyzed from the perspective of static and dynamic demand.Based on the independent analysis of the three macroscopic characteristics,we study the influence of the interaction among the three characteristics.By referring to the basic map of macro traffic in other environments,we construct the basic chart of pedestrian crossing in line with the domestic urban traffic environment.Also,using the full temporal-spatial data to calibrate the basic graph,we introduce the density as the main variable that affects the change of the basic map,and the critical density index of the phase change is given.(3)Analyzing the micro-behavior feature of pedestrian crossing based on the full temporal-spatial informationThis paper analyzes the microcosmic characteristics of pedestrian individuals and groups based on temporal-spatial information,and we construct a typical basic character description model.On the basis of macroscopic characteristics analysis,we further discuss the microcosmic characteristics such as instantaneous speed,acceleration,start-up time,space requirements,critical crossing clearance,and patience limit through data mining of the pedestrian’s temporal-spatial information and combined with the form of questionnaire survey.We study the interaction relationship between the microscopic parameters,and then consider the situation affected by the physiological and psychological aspects.In addition,the critical limit of the traversed gap and the patience time are also analyzed.We summarized the spatiotemporal characteristics of pedestrian individuals and groups,such as avoidance,transcend and following.Thus this paper achieves the goal of describing typical basic characteristics of pedestrian crossing,providing theoretical support for pedestrian organization and control,traffic signals and road settings,which lays the foundation for the following-up pedestrian crossing mechanism modeling.(4)Constructing the mechanical model of crossing behavior and the simulation verificationWe construct the model of pedestrian crossing behavior and carry on the contrastive simulation verification through hierarchical behavioral mechanisms.Through the analysis of macroscopic,microscopic characteristics and related influencing factors,and then combined with the current research results of physiology and psychology,we study the framework of pedestrian crossing mechanism and propose a hierarchical framework model of the mechanism.And then refine it and decompose pedestrian cross-street behavior process from two levels,that is perceived planning-decision implementing.Through the data analysis and induction,the spatiotemporal subjective-benefit Dijkstra model with incomplete information and optimal selection model the behavior model with perceived social force in finite space are established,which conducts a comprehensive answer to the pedestrian crossing behaviors.Finally,the corresponding simulated environment is established according to the actual situation,and the simulation test is carried out.The test results are compared to the data extracted by the real environment.These results verify the accuracy of the crossing mechanism under the condition of full temporal-spatial information.Finally,we summarize the content of this paper,and generalize the research results and innovation points.In addition,the future research of this paper is prospected. |
