An Evaluation of a Haar Cascade Classifiers using Multi-Resolution Images and Multi-Threading Resources on a Raspberry Pi

Qussay Salih

Abstract: Image processing plays a crucial role in vision-based IoT sensors, serving various applications to enhance productivity. Researchers have highlighted computational challenges in object detection on low-cost devices like the Raspberry Pi. In today's fast-paced technological landscape, the need for automated systems delivering accurate results is paramount to task completion. This study introduces an effective multithreading approach for the Support Vector Machine (SVM) method. We have implemented a multithreading algorithm for the SVM recognition processes, harnessing the power of multicore CPU utilization. Our evaluation incorporates Memory usage, CPU Temperature, FPS, Confidence levels, and Elapsed time on the Raspberry Pi platform, with the primary goal of addressing real-time computation challenges using the Pi camera. The experimental results demonstrate a notable enhancement in detection confidence, affirming that multithreading significantly bolsters detection performance on Raspberry Pi processors across various image resolutions.

Keywords: Haar Cascade Classifier, Haar-like features, Supervised Vector Machine algorithm, Real time , Raspberry Pi