ML & AI in Object Detection Research: Evaluation, Deployability and Trends (2025)

This work investigates modern object detection techniques and evaluates their real-world deployability across accuracy, latency, and energy constraints. Using benchmark datasets such as MS COCO and an ImageNet detection subset, the study evaluates 15 state-of-the-art models—including CNN-based architectures, YOLO variants, Faster R-CNN, and SSD—under a unified experimental protocol. The analysis covers CPU usage, memory consumption, throughput, energy cost per inference, and system responsiveness. Model performance varies significantly across hardware environments, with accuracy (mAP) ranging from 0.559 to 0.863, latency from 1.07s to 6.83s, and energy consumption from 69.3J to 511.8J. To assess practical suitability for real-time environments, the work introduces a normalized Deployability Score (0–5) combining accuracy, latency, and energy efficiency. The study highlights key trade-offs between computational cost and detection precision, identifies challenges in real-time deployment, and outlines future research opportunities in efficient, scalable, and edge-ready object detection. The paper provides consolidated experimental results, comparative evaluations of major architectures, and reproducible methodologies aimed at supporting further advancements in AI-driven object detection.