Counting people with support vector regression

A special and simple method is proposed to improve people counting by adopting low-leveled feature extraction and pattern predicting techniques. Firstly, we use a morphological background modeling and a Gaussian masking method to distinguish moving targets more effectively from video frames. Then, we proposed a Histogram of Oriented Gradient (HOG) feature extraction to catch more meaningful characteristics of appearance and shape of pedestrians. Other features such as edge features and texture features, are integrated as inputs to learn a support vector regression machine and finally to predict the number of pedestrians on a video frame. The experimental results indicate that our proposed method has better performance than other methods, on both database of [4] and ours.