Multiple Scaled Person Re-Identification Framework for HD Video Surveillance Application
Person re-identification is an important problem in automated video surveillance. It remains challenging in terms of extraction of reliable and distinctive features, and matching of the features under different camera views. In this paper, we propose a no
- PDF / 752,951 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 78 Downloads / 148 Views
2
Institue of Image Communication and Network Engineering, Shanghai Jiao Tong University, Shanghai 200240, China Shanghai Key Lab of Digital Media Processing and Transmission, Shanghai 200240, China {hyang,xinyuwang}@sjtu.edu.cn, [email protected], {suhangss,jizhu1023}@gmail.com
Abstract. Person re-identification is an important problem in automated video surveillance. It remains challenging in terms of extraction of reliable and distinctive features, and matching of the features under different camera views. In this paper, we propose a novel re-identification strategy for person re-identification based on multiple image scaled framework. Specifically, global features and local features are extracted separately in different image scales. These two-scaled processing are constructed in a cascaded system. We use semi-supervised SVM to obtain a similarity function for global features and a similarity function combining the spatial constraint and salience weight for local features. Experiments are conducted on two datasets: ETHZ and our dataset with high resolution. Experimental results demonstrate that the proposed method outperforms the conventional method in terms of both accuracy and efficiency. Keywords: Person re-identification Distance metrics
1
·
Multiple scaled framework
·
Introduction
Person re-identification across different views of cameras is a fundamental task in automated video surveillance.Despite best efforts have been made in computer vision area in the past years, person re-identification problem remains largely unsolved. This is due to a number of reasons. First, the resolution of the current monitored cameras is not high enough so that person verification relying upon biometrics is infeasible and unreliable. Second, as the transition time between disjoint cameras varies greatly from individual to individual with uncertainty, it is hard to impose accurate temporal and spatial constraints. Third, the visual appearance features, which are extracted mainly from the clothing and shapes of people, are intrinsically indistinctive for matching people. To solve the re-identification problem, discriminative and reliable signature for the person is needed. The image can be described by color[1], shape[2, 3], c Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part II, CCIS 547, pp. 219–228, 2015. DOI: 10.1007/978-3-662-48570-5 21
220
H. Yang et al.
texture[3, 4, 5, 6], Haar-like representations[7], edges[3], interest points[8, 9, 10] and image patches[4]. Since a single type of features is not powerful enough to capture the subtle differences of all pairs of objects, multiple features are combined here to make the person signatures more discriminative and reliable. Bazzani et al.[1] and Cheng et al. [11]combined MSCR descriptors with weighted Color Histograms, achieving state-of-the-art results on several widely used person re-identification datasets.There are also some other research works on person re-identification have been done to learn reliable and effective mid-level features. Li et al.
Data Loading...
