Research On The Video Detection Method On The Amount Of Pedestrians Using Bayesian Inference

In this paper,we research on prediction of the number of pedestrians.Dividing a image into foreground block including pedestrian in a scene.Before Bayesian program method to infer the number of foreground block,we obtain the prior knowledge in the block.Then counting the number of every block in the image.The method mainly includes the following:1)In the step of segmenting the foreground block,the foreground block was first segmented by the background model.In order to reduce the influence of shadow,we took the K nearest k-nearest neighbor method for detailed segmentation.2)During abstracting the feature at foreground block,the effect of the perspective deformation is considered.We use the self-calibration method based on the fixed target is used to generate the perspective transformation matrix to eliminate the influence of the perspective deformation.The block entropy and region covariance features,which based on the content difference and irrelevant to the block scale are calculated.These features would be better to represent the foreground block.3)Bayesian program learning get the prior knowledge first,obviously that the area features increases gradually as the number of pedestrian block increases.By building a Bayesian network map,then establish the prediction model based on Bayesian inference by small amount of samples.In the experimental section,the model established under the small number of pedestrian get a higher accuracy rate.The prediction model of pedestrian proposed in this paper that can be used to obtain the number of passengers at the station,optimize the public transport scheduling program.To detect the number of pedestrians in the entrances and exits of public places,to evacuate crowd and guide to ensure the safety.It also can be used to detect the real-time number of pedestrian of waiting area near sidewalk,in order to optimize the signal timing of pedestrian crossing.