H. Sugano, T. Shinoda, Y. Nakamura, and R. Miyamoto (Japan)
Parallel Implementation, Cell Broadband Engine, Pedestrian Tracking
Currently we must realize a pedestrian recognition sys tem based on image processing for embedded applications, such as automotive applications, surveillance, and robotics. In these applications, highly accurate recognition must be achieved with real-time processing under limited system resources. To achieve such a pedestrian recognition system, we implement a skeleton-based pedestrian tracking scheme which can track pedestrians accurately on Cell Broadband Engine (CBE), one of the latest high performance embed ded processors for general purpose use, which has a novel memory management system to achieve efficient compu tation with parallel execution units. Experimental results using a test sequence show that processing time is reduced drastically when the number of SPE used for computation increases.
Important Links:
Go Back
