Availability of low-cost and miniature size cameras, such as those already used in mobile phones, enables deployment of these devices on regular sensor nodes to extract more descriptive information about the environment. We call these Multimedia Sensor Networks (MSN). MSNs can be used for video survelliance applications for border protection, health care, etc. Multimedia data not only enhance information collection, but also improve coverage since cameras can observe everything within their line of sight, in comparison to a single discrete point observation, e.g., temperature, etc., with ordinary sensors.

In traditional multimedia applications mostly resourceful sources, e.g., servers, communicate to potentially resource-constrained receivers, e.g., mobile devices. Senders usually deploy costly encoding that would allow efficient and easy decoding at the receiver. On the other hand, in multimedia sensor networks the complexity balance is reversed; i.e., resource-constrained sensor nodes communicate with the resourceful receiver, the sink node. As a result, there is a pressing need to reverse the traditional processing complexity. It is preferable to deploy efficient processing at the source to reduce the data to transmit at the possible cost of increased processing complexity at the receiver, e.g., the sink node.

In this project, we study the major research challenges in wireless multimedia networks for surveillance video applications:

• Bandwidth Demand: Sensor nodes are constrained in terms of battery, memory, processing capability, and achievable data communication rate. On the other hand, multimedia traffic has high bandwidth demands. Therefore, it is required to reduce the amount of data that needs to be transmitted for multimedia sensor network applications. For this purpose, we apply background extraction techniques to reduce the data obtained with the camera. When possible we only report event descriptions rather than multimedia frames.
• Processing Costs: It is essential to reduce the amount of data that needs to be communicated between the sensor nodes. Yet, it is not possible to deploy complex and costly processing at the nodes due to processing constraints. Processing costs can easily become a bottleneck in sensor nodes making solutions economically unfeasible. For this purpose, we limit the overhead of background substraction and avoid costly event detection techniques.
• Storage Costs: Many multimedia processing techniques depend on historical data to enable offset descriptions from previous frames, background data descriptions, etc. In sensor networks, in addition to processing limitation on individual sensor nodes, the storage capacity is also limited in comparison to traditional counterparts. Therefore, dependence on historical data should be kept at a minimum such that storage demands do not exceed those that can be supported by individual nodes.

To reduce processing and storage costs as well as bandwidth demands, we introduce a novel framework for multimedia processing. In this framework, video and audio data are processed at the sensor nodes and then compiled at the sink node to produce possible object and event definitions.

  People:
    Demet Aksoy (UC Davis)

    Adnan Yazici (METU)
    Hakan Oztarak (METU)
    Roy George (Clark Atlanta University)
    Chi Nguyen (UC Davis)

  Publications:

• H. Oztarak, A. Yazici, D. Aksoy, R. George Multimedia Processing in Wireless Sensor Networks, in IEEE Innovations, Dubai, 2007.
• H. Oztarak, A. Yazici, and D. Aksoy, Efficient Fuzzy Multimedia Processing in Wireless Sensor Networks, submitted to the special issue on Multimedia Sensor Networks, 2007.