Benchmarking Cpu and Edge Tpu Acceleration For Real-Time Object Detection In Frigate Video Surveillance Systems

This paper presents an conducted benchmarking study of a Frigate-based intelligent video analytics system running on an Intel Core i5-11400H platform, with and without Google Coral Edge TPU acceleration, in a real deployment scenario that combines native Frigate features with custom computer vision extensions. The evaluated pipeline covers object detection for people, cars, dogs, and cats, face recognition for selected individuals, license plate recognition for known vehicles, and custom gesture recognition for thumbs-up, thumbs-down, peace, and stop-hand gestures, enabling a broader comparison between general-purpose CPU execution and edge-accelerated inference in a home or smart-environment context. The study is structured to distinguish literature-supported platform capabilities from original experimental measurements, with benchmarking focused on latency, throughput, CPU usage, RAM, storage, temperature, and energy behavior across Coral-enabled and CPU-only configurations.