A review of object detection methods: lightweight and energy-efficient detectors

Abstract

Object detection means finding and naming objects inside an image or a video. Today, making these models lightweight and energy-efficient is very important. This is because we want to run them on small devices like smartphones, drones, and smart cameras that have limited battery life and low computing power. Standard object detection models are too heavy and consume too much power because they have to check thousands of areas in an image at the same time. On the other hand, very small models sometimes make mistakes and draw inaccurate boxes around objects. To solve these issues, researchers have created many lightweight models. This paper reviews these methods in a simple, clear way. Instead of just listing past papers, we compare them using real numbers like accuracy (mAP), speed (FPS), model size (parameters), and power consumption metrics. We cover both deep neural networks (NN) and traditional non-neural network methods, explain their pros and cons, and show what is still missing in this field to help future researchers build better models. Additionally, to bridge the gap between high-level algorithmic design and low-power hardware realities, we introduce a strategic features framework (Fig. 21) and an operational, scenario-driven deployment roadmap (Fig. 22) to serve as a practical selection guide for system designers.