Stereo Matching Frameworks for Depth-Aware Object Detection: A Comprehensive Review

Abstract

Stereo matching is a fundamental technique for estimating depth from stereo image pairs, and it remains essential for object detection tasks that require accurate three-dimensional perception. This review examines classical, semi-global, and deep learning stereo frameworks, emphasizing their operational principles, strengths, and limitations. The study highlights the importance of disparity reliability for real-world applications in autonomous driving, robotics, medical imaging, agriculture, and remote sensing. Key challenges are identified, including texture ambiguity, occlusion, illumination variation, repetitive patterns, and computational burden, all of which influence the performance of stereo-based detection systems. Insights from recent literature show that advances in adaptive aggregation, transformer-based models, temporal fusion, and multi-sensor integration have improved depth stability and detection accuracy across complex environments. This review provides a consolidated understanding of stereo matching developments and outlines opportunities for designing robust, efficient, and application-aware stereo frameworks for next-generation object detectio.