NOVEL APPROACH FOR OBJECT DETECTION IN EMBEDDED SYSTEMS USING YOLO

This paper presents a comprehensive approach for implementing YOLO (You Only Look Once) object detection algorithms on resource-constrained embedded systems. With the increasing demand for real-time computer vision applications in edge devices, there is a critical need for efficient object detection methods that can operate within the computational and power limitations of embedded platforms. This research introduces novel optimization techniques including model compression, quantization strategies, and hardware acceleration methods specifically tailored for YOLO architectures. Our proposed methodology achieves significant performance improvements, reducing inference time by up to 73% while maintaining detection accuracy above 95% of the original model performance. The implementation demonstrates successful deployment on various embedded platforms including Raspberry Pi, NVIDIA Jetson devices, and ARM-based microcontrollers. Experimental results show that our optimized YOLO im- plementation can achieve real-time performance (> 30 FPS) on low-power embedded systems while consuming less than 5W of power.