Importance of Architecture Knowledge and Algorithm Redesign for Efficient DSP Programming

In the recent past, there has been an increase in the demand for vision-based embedded systems. These applications have high computational requirements and therefore severely strain the limited resources available in most embedded systems. With today's high-performance digital signal processors (DSPs) offering additional features to address multi-media needs, it is necessary to understand and exploit these features to realize a real-time vision-based application. Algorithm level optimizations also contribute to reduction in processing power and memory requirements. Hence, software redesign is an important process for optimizing the software. In this paper, we emphasize the need for an in-depth knowledge of the architecture for optimal, cost-effective real-time performance. We also explain the impact of various optimization techniques on the execution time and memory consumption of a simple block-based algorithm and a complex real-world stationary vehicle detection algorithm