English
a summary:
A wireless sensor network (WSN) consisting of nodes equipped with advanced low-cost cameras or image sensors is known as visual sensor network (VSN). The main function of a VSN is to capture images and send them to sink nodes for processing. Common applications of VSN are surveillance, tracking, crowd management, scientific research, etc. Such applications require large amounts of data to be exchanged between camera and sink nodes. Image data is much larger than common sensor data such as temperature, humidity, and pressure. For data delivery in VSNs, communication is constrained by several stringent QoS requirements such as delay, jitter, and data reliability. Moreover, due to the inherent limitations of WSN such as limited power, low CPU power, and insufficient memory; a VSN engineer must choose the appropriate architecture, image compression algorithms, and communication protocols depending on the application. In the literature, various techniques have been proposed to solve the bandwidth requirements of VSNs. Most of these techniques are based on visual data compression. This paper focuses on one such aspect, the communication protocol for VSN. In this paper, we present the Priority Routing Framework for Image Transport (PRoFIT), a novel routing framework for VSNs to deliver critical image information with system time constraints. We demonstrate that PROFIT improves the response time in various VSN applications compared to lower priority routing techniques. We have implemented PROFIT along with an image processing application using Contiki and simulated it on a Cooja simulator to support our claim.
introduction
A VSN consists of three types of nodes; optical sensor nodes, intermediate nodes, and sink nodes. The optical sensor node is equipped with a sensor that captures images. Depending on the application, this sensor can be of the type that captures multi-color images, gray-scale images, thermal images, infrared images, etc. Nodes equipped with these sensors require more power to run additional hardware and software components such as frame grabbers and image encoders. These nodes capture the raw images, encode them, and send them to the sink.
The main task of intermediate nodes is to send packets from visual sensor nodes to sink nodes.
These nodes may also participate in sensing other numerical environmental variables such as temperature, humidity, pressure, and concentration level of some chemicals depending on the nature of the VSN application. In addition, these nodes may also participate in encoding image data as a class of image encryption algorithms to offload some processing to intermediate nodes in order to conserve the energy of the optical sensor nodes.
Sink nodes are responsible for processing images captured by camera nodes. For this purpose, sink nodes are energy-rich and have high computational power. In order to take action depending on the VSN application, these nodes may additionally contain actuators or may be connected to a fourth type of node called actuator nodes.
The basic requirement of a WSN is to sense environmental factors using low-power and low-cost sensors and route the important data to energy-rich sink nodes for processing. This requirement becomes a challenge in a VSN because the amount of data to be transmitted is much larger than in a traditional WSN due to the type of data being shared. Surveillance applications require very large amounts of data to be exchanged between camera nodes and sinks. In traditional WSNs that sense light, humidity, pressure, etc., the traffic generated by a sensor node is limited to numerical data.
In most cases, the amount of memory required for storage and transmission is 16 bits per read.
On the other hand, a VSN node, equipped with a camera, generates vector data. For example, a raw RGB image of 128 x 128 pixels with 24 bits per pixel (8 bits per color) would be 128 x 128 x 24 = 393,216 bits (48 KB). This is several times larger than traditional sensor data.
To reduce the size of image data, image compression techniques such as discrete cosine transforms [4] or discrete wavelet transforms can be used. Although these algorithms reduce the size of the image, this reduction is not comparable to that of traditional sensor data. Therefore, image data compression is not sufficient. The processing power of each node is also limited. In addition, network topology and routing protocols play a crucial role in transmitting image information from visual sensor nodes to host nodes. Therefore, the tasks of capturing, compressing, and transmitting image data are some of the most challenging tasks faced by VSN engineers.
As mentioned before, using image compression algorithms, the data size can be reduced to some extent. A class of image compression algorithms creates multiple layers of compressed image data. The first layer contains the most prominent features of the image, for example, the edges of objects or the rough image data. The subsequent layers contain the details that when combined with the first layer, restore the original image. Some image processing algorithms consist of multiple passes that require different levels of details of the encoded image for each pass. By using such algorithms in VSNs, the system response time can be reduced. If the sink nodes receive the image data required for the first pass sooner than the data required for subsequent passes, they can start processing the first pass and take action accordingly while the data of subsequent layers reaches the sink node. This paper helps to alleviate the routing challenges of image processing algorithms by proposing a routing framework based on the following features:
(1)Visual sensor nodes must be able to prioritize outgoing packets. In this way, the image data of the first pass can be sent with a higher priority than the data of subsequent passes.
(2) Intermediate nodes or routing nodes must be aware of the packet priority so that packets with higher priority are forwarded before packets with lower priority.
(3) In case of collision of packets from two nodes, the high priority packets must be retransmitted before the low priority packets.
(4)Finally, in case of congestion, low priority packets should be dropped before any high priority packets are dropped.
The following section summarizes the different communication protocols used in VSN architectures as of today. Section 3 discusses the VSN application scenario that we consider in this paper. Section 4 defines PROFIT, the proposed priority-based routing framework. The implementation of PROFIT is discussed in Section 5. Simulations are performed to determine the usefulness of PROFIT. In Section 6, the simulation environment and results are discussed. Section 7 discusses the transfer of image features for real-time VSN applications using PROFIT to improve system response time. Finally, the paper is concluded in Section
VSN Application Scenario
This section describes the VSN application scenario discussed in this paper. In our scenario, the optical sensor node captures an image and divides it into bit levels. A bit level of a digital image is a set of bits corresponding to a specific bit position in each pixel value of the digital image.
For example, if a grayscale image is represented using 8 bits (1 byte) per pixel, then of these 8 bits, bit 1 represents the least significant bit while bit 8 represents the most significant bit.
PROFIT – Priority Routing Framework for Image Transport
This section provides details on how PROFIT works. PROFIT can be distributed across two layers of any protocol stack in addition to the thin application interface layer (AIL). These layers are the network layer and the medium access layer. AIL encapsulates the details of the network layer and the medium access layer. Figure 3 depicts the PROFIT layers and their functional details are given in the subsections below.
Implement PROFIT using Contiki OS
To measure the usefulness of PROFIT, a VSN implementation was created and simulations were performed. Although the framework could be implemented using any protocol stack, Contiki OS was chosen due to its acceptance as a real-time operating system for WSNs. The architecture is illustrated in Figure 5. Modifications were made to the network and MAC layer of the RIME protocol stack portion of Contiki OS. The RIME protocol stack provides a set of basic communication primitives ranging from best-effort.
Simulation results
The type of VSN applications targeted in this paper can be implemented using low-cost sensor network nodes such as TelosB [28]. Some cameras can be equipped with CMUCam4 which gives them the ability to capture images. The remaining TelosB nodes can be used to route image data from the visual sensor nodes to the nodes. These VSN applications can capture images and use image encryption algorithms such as discrete cosine transform or discrete wavelet transform to encrypt images to different levels of
The image is transferred using PROFIT.
Image features are information about an image that can be used in many image processing and computer vision applications. The first step in many image processing and computer vision applications is to extract features. These image features are edges, corners, points, SIFT, etc. Edges and corners are very basic features and these two basic features can be used in motion detection, image registration, object detection and recognition, image mosaics, and 3D modeling.
Finally:
Based on all the simulation results, we can conclude that PROFIT, the proposed priority routing framework for image transmission, helps in progressive image transmission in VSNs. Important image information can be received from visual sensor nodes in sink nodes faster than less important image information. Progressive image transmission helps us manage the low bandwidth problem of VSN environment. Priority-based transmission of bit levels using PROFIT is discussed.