Multiple feature points representation in target localization of wireless visual sensor networks

https://doi.org/10.1016/j.jnca.2015.07.013Get rights and content

Abstract

This paper discusses the target localization problem in wireless visual sensor networks. Additive noises and measurement errors will affect the accuracy of target localization when the visual nodes are equipped with low-resolution cameras. In the goal of improving the accuracy of target localization without prior knowledge of the target, each node extracts multiple feature points from images to represent the target at the sensor node level. A statistical method is presented to match the most correlated feature point pair for merging the position information of different sensor nodes at the base station. Besides, in the case that more than one target exists in the field of interest, a scheme for locating multiple targets is provided. Simulation results show that, our proposed method has desirable performance in improving the accuracy of locating single target or multiple targets. Results also show that the proposed method has a better trade-off between camera node usage and localization accuracy.

Introduction

In wireless visual sensor networks (WVSNs), sensor nodes that are equipped with cameras have functionalities of capturing visual information about targets and delivering the visual data to a base station for further analysis and decision making. Thus, WVSNs is capable of various security and surveillance applications, such as public security, facilities surveillance and monitoring. For most of these applications, the users are interested not only in existence of targets, but also in the positions of the targets (Liu et al., 2010), because the positions could facilitate target detection, recognition and tracking.

The task of localization provides with coordinates of both sensors and targets in sensor works (Soro and Heinzelman, 2009). Thus, localization task contains self-localization of sensor nodes and target localization. In this paper, we focus on the problem of target localization while the locations of sensor are already known. Target localization is to estimate the location of a target in the world coordinate based on the visual information of camera nodes (Kulkarni, 2007). The problem of target localization is well studied in wireless sensor networks. The measurement techniques in sensor localization include angle-of-arrival (AOA) measurements, distance related measurements and received signal strength (RSS) measurements (Mao et al., 2011). The existing techniques of target localization cannot be applied in WVSN. For example, multi-target can be cooperatively tracking by the Markov chain Monte Carlo data association method (Jiang and Hu, 2013). However, this method cannot address the problem of target localization in WVSN due to the significant differences in information capturing and processing between visual sensors and binary sensors. Actually, target localization in WVSN faces great challenges. Firstly, image processing is in general costly to implement in local nodes (Ercan et al., 2006), because the capabilities of computing are limited in local nodes. Secondly, the bandwidth resources are also restricted in WVSNs. Thus, there are constraints to transmit a huge amount of visual data generated by cameras to a central node or a base station (Charfi et al., 2009). Thirdly, since the sensing capability of a camera is characterized by directional sensing, the location information of a target in the depth dimension is lost in an image. Fourthly, due to the cost limitation, visual nodes in WVSN are equipped with low-resolution optical sensors (Akyildiz et al., 2007). Thus, the accuracy of filtering and extraction of target׳s position relevant information cannot be guaranteed in local sensor level.

Vision-based surveillance by multiple cameras receives considerable attentions, since visual surveillance by multiple cameras will enlarge the area and information from multiple views can be used to solve many problems (Liu et al., 2010). For example, the accuracy of the target localization can be gradually improved by selecting the most informative cameras based on correlation functions (Dai and Akyildiz, 2009) and the properties of the overlap region of the target in images (Li and Zhang, 2012) until the required accuracy level of target state is achieved. However, the multiple cameras bring new problems. Finding the correlated points pair in different images of cameras is a very difficult task. Furthermore, the energy and wireless channel capacity are very limited in WVSN. As discussed above, WVSN is a kind of resource-limited networks in nature. It is desirable to balance the trade-off between the accuracy of localization and the resources of WVSN.

The motivation of our study is to use the visual data acquired from the camera nodes to accurately estimate the position of target in the world coordinate. In this paper, we provide a method that uses multiple feature points to represent targets, and then provide a statistical approach to find the most correlated image point pair from different cameras, in order to reach the goal of improving the accuracy of target localization. Note that we focus on 2-D target localization on the ground plane. We assume that the cameras are placed horizontally around a room, which is the most relevant case for many real world applications. Besides, this paper makes the following assumptions about the wireless visual sensor network. Firstly, the location and orientation of each camera node is known within a universal coordinate system. Once a node enters into the networks, its geographical position remains constant. Next, all of the cameras are well calibrated. Finally, all of the nodes are time synchronized.

The initial results of this research have been published in Li et al. (2011), where we briefly introduced our target localization algorithm. In this paper, we expand on that work by providing further insight on the representation of the target by multiple feature points in target localization. In Li et al. (2011), we mainly focus on the single target localization. In this study, we provide a scheme for multiple target localization. The problems of the corresponding target matching and the occlusion between targets are also addressed in multiple target localization without prior knowledge. Besides, we also expend the experiments in a practical environment. Our main contributions include the following: 1) we design a method of using multiple feature points to represent a target in the goal of decreasing the influence of measurement noises and errors on the localization accuracy; 2) we proposed a statistical approach to match the most correlated feature points in different cameras; 3) we provide a scheme to address the issue of multiple target localization without prior information. The rest of this paper is organized as follows. Section 2 briefly highlights the related work. Section 3 presents the standard geometrical epipolar model to compute the possible position of single target. Section 4 proposes the technique to represent a target by multiple feature points and the method for finding the correlated point pair by the statistical method. Section 5 studies the case when there is more than one target existing in the field of interest. Section 6 conducts experiments to validate and evaluate the effectiveness of our proposed method and conclusions are given in Section 7.

Section snippets

Related work

Recently, a lot of researches focus on visual sensor networks, but very limited works related to the target localization in WVSNs has been reported.

Farrell et al. (2009) present a system that uses two cameras to localize the node of wireless sensor networks, and then employs non-imaging sensors to estimate the location of targets. Liu et al. (2010) described the common procedure of collaborative single target localization in wireless visual sensor networks, which extracts one feature point and

General target localization model

In WVSN, since multimedia content, especially video streams, requires transmission bandwidth that is orders of magnitude higher than that supported by currently available sensors (Akyildiz et al., 2007), it is crucial to perform as much local processing (such as compression, error protection, filter, feature extraction, etc.) as possible to reduce the amount of information that needs to be communicated to other nodes (Sanchez-Matamoros et al., 2009). In this study, if cameras capture any

Multiple feature points representation in target localization

In the most existing methods of target localization in WVSN, the target is regarded as a feature point after extraction in images. However, the one-feature-point extraction approach is invulnerable to the additive noise and measurement errors, especially when the cameras׳ resolutions are not good enough. If the feature point that is corrupted by noises and errors is used in determining the target׳s coordinates, the accuracy will deteriorates. In this paper, we provide an approach of the target

Scheme for multiple target localization

It is possible that there is more than one target existing in the field of interest in practical applications, e.g. battlefield and video surveillance. The biggest difference between single target localization and multiple target localization lies on the corresponding target matching. That is to say that for a target in one image, we have to find the corresponding one in the other images. Besides, we have to consider occlusions by other targets. From the perspective of statistics, the position

Evaluation on single target localization

In this section, we present some simulations to evaluate the performance of our proposed approach. All the simulations in this study have been carried out in MATLAB and Visual C++ on a Pentium machine with 2-GB RAM. For the reason of reducing the simulation complexity, we firstly deployed six camera sensors in a 10 m×10 m field. The image format is CIF and the field of view in horizontal is 57.4°. We know cameras׳ locations and projection matrices with respect to a common world frame, shown in

Conclusions

In this paper, we have provided a novel technique for locating targets in wireless visual sensor networks in the goal of improving the accuracy. This technique combines a statistical algorithm with image processing algorithms. In order to reduce the influence of noises and errors on the accuracy, multiple feature points are selected to represent the position of targets. We develop a statistical method to find the most correlated point pair to determine the target׳s position. Besides, in the

Acknowledgments

This work was supported by National Natural Science Foundation of China, No. 61403065, Project of Sichuan Science and Technology Bureau, No. 2015JY0084, Fund of Erasmus Mundus TANDAM Project, 2010 and Doctoral Fund of Ministry of Education of China, No. 20120185120034.

References (28)

  • Ian F. Akyildiz et al.

    A survey on wireless multimedia sensor networks

    Comput Netw

    (2007)
  • I.F. Akyildiz et al.

    A survey on wireless multimedia sensor networks,

    IEEE Wirel Commun

    (2007)
  • O. Barnich et al.

    ViBE: a universal background subtraction algorithm for video sequences

    IEEE Trans Image Process

    (2011)
  • Youssef Charfi et al.

    Challenging issues in visual sensor networks,

    IEEE Wirel Commun

    (2009)
  • L.C. Chiu et al.

    Fast SIFT design for real-time visual feature extraction

    IEEE Trans Image Process

    (2013)
  • R. Dai et al.

    A spatial correlation model for visual information in wireless multimedia sensor networks

    IEEE Trans Multimed

    (2009)
  • J.R.M. Dios et al.

    Chapter 3: Localization and tracking using camera-based wireless sensor networks

    (2011)
  • Ercan D Yang, Gamal AE, Guibas L. Optimal placement and selection of camera network nodes for target localization....
  • R. Farrell et al.

    Target localization in camera wireless networks,

    Pervasive Mob Comput

    (2009)
  • Funiak S, Guestrin C, Paskin M, Sukthankar R. Distributed localization of networked cameras. In: Proceedings of the...
  • W.B. Heinzelman et al.

    An application-specific protocol architecture for wireless microsensor networks

    IEEE Trans Wirel Commun

    (2002)
  • Jiang F, Hu J. Cooperative multi-target tracking in passive sensor-based networks. In: Proceedings of IEEE wireless...
  • Jin Teng et al.

    EV-Loc: integrating electronic and visual signals for accurate localization

    IEEE/ACM Trans Netw

    (2014)
  • T. Kim et al.

    Combined shape and feature-based video analysis and its application to non-rigid object tracking

    IET Image Process

    (2011)
  • Cited by (0)

    View full text