Evaluating Real-Time Pattern Recognition in Autonomous Systems on COTS Hardware

Abstract

The increasing demand for real-time performance in autonomous systems requires precise and deterministic execution of critical tasks such as object detection. This thesis evaluates the real-time capabilities of Xenomai 3, Xenomai 4, and PREEMPT-RT in enabling predictable performance for object detection tasks on the Raspberry Pi platform. The research focuses on analyzing the behavior of these real-time extensions under varying workloads and timing constraints, emphasizing their impact on latency, jitter, and system responsiveness. The study implements real-time object detection algorithms to benchmark the performance of each real-time option. Comparative experiments assess their ability to maintain strict deadlines and deliver consistent results under real-world conditions. Results are analyzed to identify the strengths, limitations, and suitability of Xenomai and PREEMPT-RT for applications requiring real-time pattern recognition and decision-making capabilities. By exploring the interplay between real-time configurations and object detection tasks, this work contributes to a deeper understanding of achieving deterministic performance in autonomous systems. The findings provide practical insights for developers and researchers aiming to optimize real-time workloads on constrained platforms like the Raspberry Pi.