Counting people in various crowed density scenes using support vector regression
多种人群密度场景下的人群计数

The use of video surveillance in for people counting public places has an important value in the field of intelligent security. However, there are several factors such as camera perspective, background clutter, and occlusions, which restrict its development and application of the study. An algorithm based on the regression model is proposed for estimating the number of people. First, in order to eliminate the effect of the camera perspective on the image features, the input image is divided into several sub-image blocks according to the change of pedestrian height in the image. Second, the simile classifier is used to improve the advanced local binary patterns (ALBP) texture feature of the blocks. Then, according to the crowd density, we use the support vector regression (SVR), which has two kernel functions to establish the relationship between input features and the number of people. Finally, adding the number of persons of all sub-image blocks gives us the total number of people on the image. Experimental results show that the absolute error of the sparse population is approximately one person using the presented algorithm and the relative error of the testing crowded population is less than 10%. This therefore demonstrates the high accuracy of this algorithm, which can be applied for people counting in video surveillance.