Human Silhouette Detection in Images using Machine Learning

Abstract

Detecting human outlines, or silhouettes, has emerged as a crucial task in the field of machine learning, with important applications in areas such as surveillance, human–computer interaction, healthcare monitoring, and autonomous navigation. Accurate silhouette detection is essential not only for ensuring safety but also for improving accessibility and user experience in systems designed to assist individuals in real-world environments. Unlike traditional computer vision techniques that rely on hand-crafted rules, modern machine learning models—particularly convolutional neural networks (CNNs)—are capable of learning visual patterns from data, making them more effective in handling complex and cluttered scenes. This research introduces a practical machine learning framework for human silhouette detection, focusing on identifying individuals in natural environments where backgrounds may include occlusions from trees, rocks, and other visual distractions. A custom dataset was created for this purpose, containing annotated images that reflect diverse backgrounds, human postures, and varying levels of visibility. This dataset supports realistic model training and evaluation by simulating challenging real-world scenarios. The study employs and compares two deep learning architectures: YOLOv8n (You Only Look Once) and DETR (Detection Transformer). YOLOv8n is a lightweight, real-time object detection model optimized for high-speed performance, making it suitable for deployment in resource-constrained systems such as drones or embedded devices. In contrast, DETR applies transformer-based attention mechanisms to capture global context within an image, offering improved detection performance in scenes with overlapping or partially occluded human figures. By evaluating both models on the custom dataset, the research highlights their relative strengths in terms of accuracy, speed, and suitability for different deployment scenarios. Overall, the work outlines a structured approach to designing and evaluating human silhouette detection systems using state-of-the-art models and a dataset tailored to real-world conditions. The findings contribute to a better understanding of the trade-offs involved in deploying deep learning-based detection systems and provide a foundation for further development of reliable and adaptable computer vision solutions.