People counting based on improved gauss process regression

Abstract

Ideally, in the method about people counting based on multi-feature regression, the features, such as weighted blob area and perimeter, should have a linear relationship with the number of people in the scene. However, although the overall linear trend, due to the existence of occlusion, the foreground extraction errors and other factors, the local presents nonlinear characteristics. Gauss process regression is very suitable for linear features with local nonlinearity, so it is widely used at present to achieve the regression analysis between the features and the number of people using the Gauss process regression. In order to obtain higher accuracy, based on the research of the insufficient of the traditional Gauss process regression method, an improved Gauss process regression method is proposed to people counting. The experimental results show that the proposed method can get better performance. Firstly, the foreground blob and features of image sequences are extracted. Next, the square exponential covariance function is selected as kernel function. The bacterial foraging algorithm is used to optimize the hyper-parameters to obtain the optimal solution, and then the regression model is established. The experimental results show that the proposed algorithm which makes use of bacterial foraging to optimize the hyper-parameters can obtain better parameters and improve the accuracy of the people counting.