Abstract

In computer vision, object detection is a core challenge, particularly in scenarios requiring real-time responses and limited computational resources. SSD is a single-pass object detection method that combines classification and localization, recognized for its rapid processing and efficiency. This systematic review analyzes the application and enhancement of SSD-based models across various domains between 2020 and 2025. The study focuses on three major challenges in object detection: real- time performance, small object detection, and deployment in limited-resource settings. It examines diverse strategies, including domain-specific data augmentation, architectural modifications such as deformable SSDs and DRMP modules, and the integration of lightweight backbones like MobileNet. The review highlights significant performance gains in accuracy and efficiency but also reveals persistent limitations, including difficulties in detecting small objects in complex scenes, vulnerability to occlusion, and challenges in model generalization. This study offers key insights into recent research trends and future directions for optimizing SSD-based frameworks to meet the demands of increasingly complex and dynamic detection scenarios.

Concepts :

Access to Document

Citations by Year