An efficient framework for counting pedestrians crossing a line using low-cost devices: the benefits of distilling the k
- PDF / 2,347,003 Bytes
- 15 Pages / 439.642 x 666.49 pts Page_size
- 85 Downloads / 143 Views
An efficient framework for counting pedestrians crossing a line using low-cost devices: the benefits of distilling the knowledge in a neural network Yih–Kai Lin1 · Chu–Fu Wang1 · Ching-Yu Chang1 · Hao–Lun Sun1 Received: 14 August 2019 / Revised: 3 June 2020 / Accepted: 26 June 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract In this paper, an efficient framework for counting pedestrians crossing a line of interest is proposed. Nowadays, the convolutional neural networks have very good results on pedestrian detection and tracking. However, the major drawback of the neural networks is that they require heavy computing resources. This limits the application of neural networks in low-cost systems. Thus, the low power consuming pedestrian counting systems with comparable performance are still important. To achieve this goal, the proposed method distils the pedestrian detection knowledge from a neural network to train the local binary patterns (LBP) cascade classifier model to detect pedestrians. Then a matching and tracking algorithm is used to count the number of pedestrians. An automaton was developed to eliminate the bouncing position of the detected pedestrians. The experimental comparisons show that, compared to Ma et al. and Felzenszwalb et al.’s methods, the quality of the line of interest counting of the proposed method is about the same and, at the same time, the execution time of the proposed method is much less. Keywords Counting pedestrians · Distilling knowledge · LBP cascade classifier
1 Introduction An important resource management issue in public places, e.g., museums, train stations, shopping malls etc., is the number of visitors in a region. The goal of knowing how many visitors there are can be reached by counting the number of pedestrians entering and exiting, which can be achieved by counting the number of pedestrians passing a line of interest (LOI). The LOI counting problem, the purpose of this paper, is to count the number of pedestrians crossing the pre-defined lines from the video captured by surveillance. Three Supported by the Ministry of Science and Technology of Taiwan under contracts MOST-108-2221-E-153-005 Yih–Kai Lin
[email protected] 1
National PingTung University, No.4-18, Minsheng Rd., Pingtung, Taiwan
Multimedia Tools and Applications
main difficulties of the LOI counting problem are: 1) the angle of video surveillance affects the perspective of the scene and causes the size and shape of objects to change dramatically; 2) in a crowd, those who pass the camera block those behind them; and 3) objects in different directions cross each other and some objects are obscured. Traditional LOI counting approaches can be divided into two categories: 1) counting crowd blobs from video [2, 3, 6] and 2) estimating the number of pedestrians in a temporal slice image by the regression method or local histogram-of-oriented-gradients (LHOG) [1, 14, 24]. The major drawback of these methods is that they require heavy computing resources. This limits the ap
Data Loading...
