Distributed calibration and tracking with low-power image sensors

摘要

The replication of a single-camera system is not in itself a solution to large scale surveillance. A scalable solution to surveillance can be achieved through employing wireless sensor network technology where distributed sensors embedded with processors communicate wirelessly. To address the constraints of unreliable communications, power-intensive processing and communications, and limited memory, DISCERN (DIStributed Camera Event Recognition Network) distributes reasoning about its organization and detected target information. This enables sensor nodes to collaborate intelligently with one another, forewarning neighboring nodes of possible targets, resolving location ambiguities of the sensor and target, and providing greater intelligence as additional target data is collected. We addressed the limited power and processing speed by incorporating low-power image processing techniques to quickly reduce the large data acquired through images. Robustness was maintained through decision-based fusionfor target detection and data-based fusion for target extraction and tracking across the sensor field. Distributed control is possible through our information-based neighbor lists, facilitating the transformation of the target's information across sensor nodes as it traverses to the end user.