ratory 9, Sha
Received 5 March 2014
Received in revised form
Human behavior analysis
Near-infrared n is a especially in applications such as visual surveillance, robotics, and drive-assistance systems. Recently, most pedestrian detection approaches of machine learning and signal the the sliding window or not . are still some n and inclusion. oding the entire whole human is tion and various human gestures, the relation between entire human and are taken into account. In addition to holistic template
Contents lists available at ScienceDirect journal homepage: www.els
Signal Processing ] (]]]]) ]]]–]]]0165-1684/& 2014 Elsevier B.V. All rights reserved.detector, part-based detectors are also explored. Discriminative part based approaches are proposed in [11,12] which can automatically decide the position of parts http://dx.doi.org/10.1016/j.sigpro.2014.08.003 n Corresponding author.
E-mail address: email@example.com (X. Lu).Pleas 10.10A series of pedestrian detection approaches can benefit from low level features and training approaches. In , body parts has been considered in [9–12]. In these works, human body parts, such as head, arm, upper body and leg,sis. Pedestrian detection endows machines with the ability to interact with human via techniques of machine learning and signal processing, and recently many algorithms for pedestrian detection have been proposed [1,3–5]. Most popular pedestrian detection approaches are template based approaches which adopt a pre-trained binary classifier to determine whether there is a pedestrian within shown that HOG feature outperforms o pedestrian detection. However, there challenges such as illumination variatio
For example, the holistic templates enc human body are effective only when the visible.
In order to handle the occlusion situacomponents in human pose recovery and behavior analy- is proposed. And the work in  has experimentally ther features inLatent SVM 1. Introduction
Pedestrian detection is one ofe cite this article as: Y. Yuan, et al., 16/j.sigpro.2014.08.003ithere exists a limitation on the accuracy in pedestrian detection. The reason behind this is that supporting information for detecting pedestrian is limited. In fact, spectrum besides visible light can provide abundant discriminative information for pedestrian detection.
Therefore, it is significative to exploit multi-spectral information for detection task. In this paper, a multi-spectral based pedestrian detection approach is proposed, which not only takes use of the information of red, green and blue (RGB) bands, but also incorporates the information of near-infrared spectrum into the detection process. Latent variable support vector machines (L-SVM) are employed to train the multi-spectral pedestrian detection model. Experiments are implemented on a new dataset containing 1826 multi-spectral image pairs. The experimental results illustrate that utilizing multi-spectral information achieves significant performance improvement in a pedestrian detection task compared with only using RGB information. & 2014 Elsevier B.V. All rights reserved. most important
Haar feature and boosting classifier is adopted. In , inspired by scale-invariant feature transform (SIFT) , histogram of oriented gradient (HOG) feature for detection24 June 2014
Accepted 4 August 2014 processing have achieved advanced performance in traditional natural images. However,Multi-spectral pedestrian detection
Yuan Yuan, Xiaoqiang Lu n, Xiao Chen
Center for OPTical IMagery Analysis and Learning (OPTIMAL), State Key Labo of Optics and Precision Mechanics, Chinese Academy of Sciences, Xi'an 71011 a r t i c l e i n f o
Article history: a b s t r a c t
Pedestrian detectioMulti-spectral pedestrof Transient Optics and Photonics, Xi'an Institute anxi, PR China crucial problem in human pose recovery and behavior analysis, evier.com/locate/sigpro cessingian detection, Signal Processing (2014), http://dx.doi.org/
Y. Yuan et al. / Signal Processing ] (]]]]) ]]]–]]]2without part-level supervision by regarding unknown part positions as latent variables in a support vector machines (SVM) framework.
Recently, pedestrian detection has been a hot research subject for visible spectrum images. Some researchers transfer the approaches developed for visible spectrum images to other band such as infrared [13,14]. The work in  has demonstrated that far-infrared (7–14 μm) can enhance the performance of human detection by employing the approach in . However, these approaches fail to utilize information of more than one spectral band.
It would lead to some limitations due to the lack of information from other spectral band. Recent researches of multiview learning have demonstrated that integrating multiple features into training procedure could improve the performance in many applications such as image retrieval , image reranking , and image classification . These approaches prove that utilizing different kinds of information can provide a refreshing perspective to solve problems of computer vision. With this consideration, multi-spectral information is provided as a new view to learn the world more effectively.
Recent researches have demonstrated the effective using of multi-spectral information in computer vision applications. The work in  has proved that nearinfrared (NIR) has a weaker dependence on red, green and blue (RGB) bands than they do to each other. Accordingly, it presents that multi-spectral color SIFT descriptor can effectively utilize multi-spectral information to enhance the performance of recognition and classification. In , near-infrared and visible cues are fused for skin enhancement. And in [17,27], near-infrared information is integrated into the framework of saliency detection, which achieves a great success. Inspired by these works on the utility of multi-spectrum, a novel idea for multi-spectral pedestrian detection is explored in this paper.