Research On People Estimating In Designated Area Based On Neural Network

This method can count the number of people in a certain area in the video. Withthe continuous development of the society, the needs of the construction ofmodernization, people counting in video have been applied in various public placesand road entrances. However, due to the complexity of the applying scenarios and thediversity of the objects in the video, there are a lot of difficulties to study, both in thedomestic and abroad researchers have proposed a variety of solutions, but there arestill many deficiencies need to be improved.This method uses the neural network to estimate the number of persons in thevideo. Firstly we get the image of designated area of the image and use the Gaussmixture model to model background, and then extract the foreground from the image,and get the feature vectors of the foreground, like the number of foreground pixelsand the number of edge pixels, and then combine with the actual number formartificial statistics, to train the neural network model. In the estimating stage, we firstextract the density information of the scene using the SURF algorithm, as thepopulation density is different; we use different parameters to adjust the extractedfeature vectors as the input of neural network model, the output is the estimatednumber in the scene. At the same time, we used the HOG human detection method toreplace the artificial statistics for counting the actual number of persons. Finally, wedo experiments in different situations, to validate the time complexity, accuracy androbustness of our method.