Researchers at the University of Michigan have developed a low-power CMOS imager with capabilities of feature extraction and object identification for use in distributed sensor networks. Wireless sensor nodes typically operate under constraints of limited energy, and power consumption is an important factor in extending their lifetimes.
Imagers used in wireless sensor networks account for a small portion of overall power consumption (under 10 microwatts) and can be powered by energy harvesting. But a very large source of energy consumption in sensor networks stems from wireless signal transmission of large bandwidth image signals. Substantial reductions in transmission bandwidth can be realized by generating signals only upon the occurrence of certain events such as temporal or contrast changes. However, event-based imaging can suffer from redundancies due illumination changes or background movement not related to the objects of interest.
The low-power CMOS image sensor system developed at the University of Michigan overcomes the above challenges. The sensor is motion-triggered and achieves object-of-interest imaging while suppressing the redundancies in both imaging and the transmission of signals. The image sensor is typically in sleep mode until woken up when triggered by motion. The sensor then extracts features from the captured image to detect objects of interest. Full image capture is performed only when target objects are found, thereby significantly reducing power consumption.
The design leverages techniques such as histograms-of-oriented-gradients for feature extraction, and makes use of mixed signal circuits to reduce energy consumption. The chip consumes 0.22 microwatts/frame for motion sensing and 3.4 microwatts/frame for feature extraction.
- Wireless distributed sensor networks for surveillance, environmental monitoring, traffic management
- Wireless biomedical imaging systems for disposable microscopes and endoscopes
- Automotive sensors for pedestrian detection
- Low power
- Reduced bandwidth
- Scalable spatial resolution