Crowd Evacuation Path Planning And Application Based On Image Feature Extractio

With the rapid development of society and economy,people’s living standards have been greatly improved,while people’s material and cultural life have been greatly enriched as well.Nevertheless,the public safety problem of crowd evacuation,caused by high-density and unstable flow of people in relatively closed spaces has attracted more and more attention from scholars.It is a hot topic to develop efficient evacuation route plans and reasonable crowd evacuation models,and the related algorithms.It is of vital scientific and practical value to carry out an extensive and in-depth study on crowd evacuation movement.In the thesis,the main work concentrates on probing into crowd evacuation path planning models and optimization algorithms in order to handle the problem of crowd evacuation path planning in static and dynamic scenarios from the perspective of path planning model design,deep learning,and intelligent optimization.The acquired results not only help to promote the development of crowd evacuation path planning research,but also provide referable evacuation solutions for solving crowd evacuation problems.The main work and the results obtained are summarized below.A.Related to the shortcomings of the classical YOLOv3 with the aspects of pedestrian detection leakage and low detection accuracy,an improved YOLOv3 is proposed to cope with the problem of overlapping pedestrian detection in dense scenes,depending on the channel attention mechanism and the spatial pyramid model.In the design of the model,the channel attention mechanism is incorporated into the residual network to reinforce the feature extraction of crucial information;the spatial pyramid model,which can enlarge the perceptual field,is used to fuse multi-scale feature maps in order to strengthen the ability of feature extraction of mutually occluded targets;the improved non-extreme suppression module is utilized to eliminate redundant prediction frames in order to avoid the missed detection of overlapping objectives.Numerically comparative experiments show that the improved YOLOv3 can effectively avoid the phenomenon of missed detection with high detection accuracy and recognition rate when applied to the detection of pedestrians in dense scenes.B.For the problem of crowd evacuation path planning in static environment,an improved social force crowd evacuation path planning model is developed,in which the evacuation time is taken as the performance index.An improved butterfly optimization algorithm is developed to solve such a problem,where the Tent chaotic mapping and an adaptive weighted nonlinear inertia weighting strategy are introduced into the basic butterfly optimization algorithm to strengthen the diversity of the butterfly population and the ability of global solution search.Numerically comparative experiments show that the improved social force crowd evacuation path planning model is available;the improved algorithm has significant advantages with the aspects of solution search stability,search effect,and convergence speed,while the acquired crowd evacuation scheduling scheme is rational and effective.C.For the problem of crowd evacuation path planning in dynamic environments,a dynamic planning evacuation crowd scheduling model is developed after extending the above improved social force crowd evacuation path planning model.Further,a dynamic optimization algorithm is designed to solve such a dynamic model,after two strategies of environmental detection and adaptive dynamic odor absorption and a dynamic switching probability mechanism are embedded to the above improved butterfly optimization algorithm.Numerically comparative experiments show that the acquired dynamic evacuation model is rational and the dynamic optimization algorithm can effectively detect changes in the dynamic environment.In addition,the algorithm can effectively obtain a high-quality crowd evacuation scheduling scheme in the dynamic environment.