A Survey of Ranging Algorithms Based on Binocular Stereo Vision

With the rapid development of computer vision and information technology, binocular stereo vision ranging has become an indispensable technology for modern science and technology due to its advantages of high precision, low cost and easy integration, and is widely used in the fields of robotics, autonomous driving, augmented reality, industrial automation and medical imaging. This paper presents an overview of the basic principles and recent advancements in binocular stereo vision ranging. It reviews traditional algorithms such as SAD (Sum of Absolute Differences), SSD (Single Shot Detector), NCC (Normalized Cross-Correlation), Census transform, SGBM (Semi-Global Block Matching), and ELAS (Efficient Large-Scale Stereo Matching). These algorithms are known for their computational efficiency and real-time capabilities, making them suitable for applications like industrial automation and real-time monitoring systems that tolerate low computational complexity and illumination variations. Furthermore, the paper examines recent developments in deep learning-based binocular vision stereo ranging algorithms, including GC-Net (Geometry and Context Network), PSM-Net (Pyramid Stereo Matching Network), Stereo-Net, and MC-CNN (Multi-Scale Convolutional Neural Network). These approaches leverage deep learning architectures to process multi-scale information, significantly enhancing stereo matching accuracy and robustness. They excel in complex scenarios demanding high-precision depth maps, such as autonomous driving and virtual reality environment reconstruction. Despite advancements, challenges remain in enhancing real-time performance, accuracy, and robustness of binocular stereo vision ranging. Therefore, further advancements are crucial, involving algorithm optimization and integration of multi-modal information to foster broader applications and technological evolution.