The USC Reduced Complexity Vision system was designed as an alternative to high priced vision systems which have traditionally cost tens of thousands of dollars. The RCV was built for under $1000, and is capable of performing low-level visual computations at near frame rate.
This is done with a concept called dynamic foveation, which allows the user to specify exactly which portion of an image is to be digitized and processed. Therefore, a sparse sampling could be taken of the entire frame. When an area of interest is noticed, the system can dynamically reconfigure it's sampling strategy to zoom into the significant area, providing greater detail, while limiting the actual number of pixels processed to 4K.
Because the number of pixels is limited to 4K, processing the images is not very computationally expensive. This allowed us to use a 68332 BCC, which is a $250 single board microcontroller based on the 68000 series.
We have demonstrated tracking of an object at 15 frames/second, and we will be using the system on the USC Autonomous Flying Vehicle.
We have a set of sample dynamically zoomed images and one of random sampling.
Batavia, P. H., Lewis, M. A., Bekey, G. , A Reduced Complexity Vision System for Autonomous Helicopter Navigation, Proceedings of the 1995 IEEE Robotics and Automation Conference, Nagoya, Japan, 1995.
A Low Cost Reduced Complexity Vision System for Autonomous Robotics and cover, Institute for Robotics and Intelligent Systems Tech Report #IRIS-94-323, 1995