This proposal focuses on the use of biomimetic image sensors for an image sensor network. Custom image sensors will provide a reduced visual representation and enable the sensor network to interpret behavior and activity in an indoor setting. The research emphasis is to 1) parse out sensor observations into a set of distinguishable actions that can be understood by machines, 2) preserve privacy by extracting data from a scene without acquiring images and by making image reconstruction difficult for perpetrators, 3) provide ultra-low power components and a lightweight sensor network of images that operates using symbolic information over low bandwidth wireless links.
Intellectual Merit: The research enables image parsing and interpretation on sensor networks. A new generation of probabilistic detection algorithms that operate on prioritized data to produce fast detections with limited information will be developed. The algorithms will be able to execute on the low-end microcontrollers used on sensor nodes. Using modular and biomimetic approaches, the PIs will design low-power image sensors that will perform efficient data extraction at the sensor level. Their design approach will reduce the computation and communications required by image sensors and provide an orthogonal methodology to conventional machine vision. The proposed research is required to advance the technology of mobile cameras for monitoring indoor assisted-living environments.
Broader Impacts: The image sensor network is the building block of building security, wide-area security, national security, monitoring health of civil structures as bridges and highways, observing natural environments and habitats, detecting severe natural events. The proposed image sensor network will be used in an assisted-living environment to monitor aging patients during everyday activity. The applied nature of the project is an attractive way to train students, particularly women and under-represented groups, and to motivate them to pursue careers in research and academia.
