Research On The Recognition And Simulation Of Collective Behavior

Collective behavior refers to macroscopic patterns consisted of continuous and ordered agents,which exists widely in diverse crowd systems,such as bacterial colony,animals group,human crowd and traffic flow.The recognition and simulation of collective behavior are the important branch of the computer vision and computer graphics,which are significant for public safety,intelligent traffic,movie,games and architectural design.The research of the recognition and simulation is an organic whole.The recognized patterns can provide a basis for the crowd motion in simulation.The simulation results can verify the effectiveness of the patterns.The main task of recognizing collective behavior is to mining the coherent patterns consisted of highly coherent tracklets according to the extracted features from crowd motion in the videos.There are still several challenges for recognizing collective behavior.Collective behavior involves the local and global motion patterns,where the varying densities and arbitrary shapes are salient characteristic.Meanwhile,the global coherent motion with complex interaction requires more accurate measurement of local coherency and analysis of global continuity.Further study demonstrates that the motion descriptor and similarity measurement are still limited to finding the latent relativity among tracklet points under the circumstances of perspective distortion and large spatial gap.The aim of simulating collective behavior is to modeling the collective decision during the crowd movement by combining the subjective and objective factors,and rendering the large scale crowd realistically and plausibly.Especially for the complex environment,the phenomenon of path follow is a significant ingredient to simulating collective behavior.However,existing works neglect the relation between human's subjective factors and collective decision in path planning,which are limited to generating the simulation in a plausible manner.This thesis focuses on the research of recognition and simulation of collective behavior.The main contributions are summarized as follows.(1)To recognize both local and global coherent motion having arbitrary shapes and varying densities,we proposed the collective density clustering method.The collective density is firstly defined to reveal the underlying patterns with varying levels of density.Based on collective density,the collective clustering algorithm is further presented to recognize the local consistency,where density-based clustering is more adaptive to recognize clusters with arbitrary shapes.Finally,the collective merging algorithm is introduced to fully characterize the global continuity.Experiments on diverse crowd scenes and the comparisons with previous works demonstrate the effectiveness of the proposed method.(2)Aimed to recognize the collective behavior with complex interaction,we proposed the dynamical kernel density based multiple interaction coherent motion detection method.We creatively define a collective density to discover underlying ordered density estimation,and subsequently a novel collective clustering algorithm is introduced,which is able to identify collective subgroups rapidly.Considering the complex interaction among subgroups,we present a hierarchical Union-Find based collective merging algorithm used to recognize coherent motion by merging collective subgroups.Experimental results on several challenging video datasets demonstrate that the proposed method achieves better results than state-of-the-art works.The proposed framework shows potential to be further applied in other problems,e.g.affine motion segmentation.(3)Existing works are limited to finding the latent relativity among tracklet points under the circumstances of perspective distortion and large spatial gap.To address this problem,we proposed a coherent motion detection method based on coherent covariance.A novel coherent covariance descriptor is proposed as motion feature instead of Euclidean feature vector,which strengthens the topological relations among neighboring points.The similarity between coherent covariances is measured on Riemannian manifolds,which is beneficial for finding latent relativity among tracklet points.Then we furnish a novel spatial weighting density-based clustering to cluster tracklets according to the similarity into different motion patterns without additional merging process.Experimental results on several challenging video datasets demonstrate that the proposed method achieves better results than state-of-the-art works,especially under the circumstances of perspective distortion or large spatial gap.(4)To address the problem that existing works ignore the relation between human's subjective factors and collective decision in path planning,we proposed a novel Emotion Contagion Model to simulate crowd path planning in dynamical environment.On the basis of OCEAN personality traits,the emotion preference is defined to bridge the gap between emotion states and path choices.Considering the agents within a collective group always have more interaction,we present an emotion contagion algorithm to reveal the dynamical variation of emotion preference under complex environment.To solve the emotion-to-path mapping,we introduce a leastexpected-time objective function to find the optimal path choice for each agent according to the navigation graph of the environment.We perform the proposed ECM on several simulation scenarios.Compared to previous works,our method can simulate the crowd path planning in a realistic and plausible manner.Further experiments by combining with current local collision-avoidance methods demonstrate the robustness of our method.