A Distributed Rate-Control Approach to Reduce Communication Burdens in VSNs

In visual sensor networks, the analyze-then-compress paradigm, where each camera process data and extract local features, is proved to be an efficient approach to reduce the amount of transmitted information. The bitrate can be further reduced by efficiently compressing the extracted features using a distributed feature coding technique. However, since the rate control is performed at the decoder, an abundant use of the feedback channel is needed to adjust the coding rate. Moreover, transmitting all extracted features, including irrelevant ones with no further contribution to the application accuracy, overloads the network. In this paper, we propose a novel feature selection and distributed coding rate control strategies that cope with these issues. The proposed strategies are designed to significantly reduce the transmitted bitrate and the communication burden with the sink, which implicitly reduces the energy consumption and the decoding delay. We show that, wisely selecting at the camera sensors level only the features effectively contributing to the application accuracy reduces the amount of transmitted information up to 34% while preserving accuracy. Furthermore, the cameras can collaborate periodically, by exchanging small amount of information about their selected features, to estimate the minimum transmission rate required for each feature based on a linear fitting model that takes into consideration the inter-camera correlation and the channel conditions. Significant average bitrate savings, reaching up to 37.71%, are achieved.