Parallelization Techniques for Spatial-Temporal Occupancy Maps from Multiple Video Streams
We describe and analyze several techniques to parallelize a novel algorithm that fuses intensity data from multiple video cameras to create a spatial-temporal occupancy map. Instead of tracking objects, the algorithm operates by recognizing free space. The brevity of operations in the algorithm allows a dense spatial occupancy map to be temporally computed at real-time video rates. Since each input image pixel is processed independently, we demonstrate parallel implementations that achieve nearly ideal speedup on a four processor shared memory architecture. Potential applications include surveillance, robotics, virtual reality, and manufacturing environments.
KeywordsLookup Table Parallelization Technique Multiprocessor Architecture Processor Workload Input Image Pixel
Unable to display preview. Download preview PDF.