Abnormal Crowd Behavior Detection Based On Crowd Speedand Crowd Distributon

In the recent years, researches in the computer vision have made many greatprogresses in various aspects. As one of the most challenging tasks in computervision, abnormal crowd behavior detection plays an important role intransportation and public safety, and has also attracted a lot of attention. Thecrowd behaviors contain the normal crowd behaviors and abnormal crowdbehaviors. Since the number of abnormal crowd behaviors is far smaller than thenumber of normal crowd behaviors, abnormal crowd behavior detection is anunbalanced problem. Because Gaussian Mixture Model (GMM) usually dealswell with the unbalanced problem, so we use the GMM over the normal crowdbehaviors to help us to predict the abnormal crowd behaviors. Since the crowddistribution information and crowd speed information of the normal crowdbehaviors and abnormal crowd behaviors are quite different, these twoinformation are used to estimate the parameters of GMM over the normal crowdbehaviors.Since the crowd distribution information is crucial for abnormal behaviorsdetection in crowd scenes, we firstly propose particle entropy approach toeffectively and accurately represent the crowd distribution information in crowdscenes. The proposed approach not only avoids unstable foreground extraction,but also owns low computational complexity. Experiments conducted onpublicly available dataset consisting of gathering and dispersion events validatethat the particle entropy can preeminently reflect the crowd distributioninformation.The particle entropy approach was proposed to portray the crowddistribution and the optical flow was introduced to describe the crowd speed.Then the crowd distribution and crowd speed are used to model the normalcrowd behaviors and predict the abnormal crowd behaviors.Experimentsconducted on publicly available dataset demonstrate that our method has anexcellent performance and the comparisons show the proposed methodoutperforms the state-of-the-art methods.