An Improved Method of Crowd Counting Based on Regression

Abstract

Recently intelligent crowd counting has attracted researchers' attention in computer vision and related fields [1][2][3][4][5][6][7][8][9]. The existing predominant techniques for crowd counting fall into two categories: 1) object detection and tracking based crowd counting; 2) crowd density estimation based on features and regression analysis.In the first category work always involves pedestrian detection and tracking. Since the counting results depend heavily upon detection response, applying a state-of-the-art detector has becoming an essential key to improve system performance. In [1-4] generic detectors including HOG based headshoulder and JRoG based part detectors have been validated fairly effective in detecting and counting people. This kind of methods shows excellent performance in scenes with sparse crowd due to its accurate localization. However, as the crowd becomes large and heavy occlusion arises, individual detection and tracking both become almost impossible. Methods in the second category, such as [5][6][7], estimate the crowd density by extracting a set of holistic or local features in regions of interest (ROI) and then modeling the number of people based on features. This kind of methods does not consider individual localization and has been proved effective in dense crowd counting, especially in public environments where heavy occlusions happen frequently. In [8], the feature-regression based system is improved by applying human template matching. Antoni B. and Vasconcelos have shown in [9] that regression based crowd estimates are substantially more accurate than those produced by state-of-theart pedestrian detectors. Thus they are more generally used in crowd counting systems. and Yanyun ZhaoAbstr act An improved method of crowd counting based on regression is proposed to support intelligent management over crowd in video surveillance systems. According to the fact that human body has an articulate structure and complicated contours of shape, we propose a new low-level feature, the number of corner points, to highlight the describable capability of the feature set. We then introduce relevance vector regression (RVR) to model the correspondence between features and the pedestrian number, and propose a fusion scheme of RVR and Gaussian process regression (GPR) to further advance the performance of the proposed algorithm. Experimental results on two crowd datasets (one is UCSDpeds) demonstrate that the proposed work outperforms state-of-the-art methods and can fulfill the real-time requirement.

An adaptive people counting system with dynamic features selection and occlusion handling

“…In high level occlusion scenarios, the relationship between the features and the number of people follows a linear trend roughly while the data fluctuates non-linearly due to occlusion [52]. A combination of linear and radial basis function (RBF) kernels are used in a high-occlusion regression model.…”

“…A combination of linear and radial basis function (RBF) kernels are used in a high-occlusion regression model. The linear kernel can capture the linear main trend well and the RBF kernel can be used to model the fluctuation of the data points [52]. Mathematically, a combination of linear and RBF kernels is given by [37], [50]:…”