Real-time Probabilistic Inference Using Approximate Computing on High Velocity Streaming Data

Andreas G. Andreou,

Poster

We will do a real-time demonstration of a real-time probabilistic processing pipeline running on four field programmable gate arrays for high velocity sensor data from an autonomous real-time ubiquitous surveillance imaging system. High performance and high throughput is achieved through approximate computing and fixed point arithmetic in a variable precision (6 bits to 18 bits) architecture. A key processing computational structure is the algorithm for change point detection, implementing exact Bayesian inference, with on-line learning for background-foreground segmentation.