立体匹配算法双目散斑体表检测系统设计-《中国医学物理学杂志》

文章信息/Info

Title:: Development of a binocular speckle surface detection system with an improved stereo matching algorithm

作者:: 王旋旋1; 2; 3; 陈永华2; 3; 丁开忠2; 3; 魏江华2; 3; 1.安徽大学物质科学与信息技术研究院，安徽合肥 230601；2.中国科学院合肥物质科学研究院，安徽合肥 230031；3.合肥中科离子医学技术装备有限公司，安徽合肥 230601

Author(s):: WANG Xuanxuan1; 2; 3; CHEN Yonghua2; 3; DING Kaizhong2; 3; WEI Jianghua2; 3; 1. Institute of Physical Science and Information Technology, Anhui University, Hefei 230601, China 2. Hefei Institute of Physical Science, Chinese Academy of Sciences, Hefei 230031, China 3. Hefei CAS Ion Medical and Technical Devices Co., Ltd., Hefei 230601, China

Keywords:: Keywords: binocular vision depth computation baseline precision system design

摘要:: 质子治疗在对肿瘤靶区精准照射时，高度依赖于患者体表的实时、高精度定位。然而，当前国内主流双目视觉检测系统在临床应用中面临一项关键挑战：在保证亚毫米级精度的前提下，难以兼顾远距离检测需求，存在“高精度则距离短，长距离则精度低”的突出矛盾。为此，本文开发一种长距离高精度的双目散斑光学体表检测系统。该系统通过硬件平台与软件算法的协同优化设计，实现在较大工作距离内保持高精度测量的目标。在硬件上，通过系统化建模与误差分析，确定最优基线长度。在软件上，提出映射矩阵压缩技术以优化立体匹配算法，大幅缩减视差搜索范围，减少图像处理需要的数据量，大幅度提高计算效率。实验数据表明，该系统实现长距离下的亚毫米级精度，性能稳定，满足质子治疗系统体表定位与呼吸监测需求。

Abstract:: Abstract: Proton therapy heavily relies on real-time, high-precision localization of the patients body surface for precise irradiation of tumor targets. However, the existing mainstream domestic binocular vision detection systems encounter a key challenge in clinical applications: difficulty in realizing long-distance detection while ensuring sub-millimeter accuracy, which results in a prominent contradiction of "high accuracy at short distances and low accuracy at long distances". Therefore, a long-distance, high-precision binocular speckle optical body surface detection system is developed, which enables high-precision measurement over a large working distance through collaborative optimization design of hardware platforms and software algorithms. In terms of hardware, the optimal baseline length is determined via systematic modeling and error analysis. For software design, a mapping matrix compression technique is proposed to optimize the stereo matching algorithm, significantly narrowing disparity search range, reducing the amount of data required for image processing, and greatly improving computational efficiency. Experimental data show that the proposed system achieves sub-millimeter accuracy at long distances, with stable performance, meeting the requirements of body surface localization and respiratory monitoring for proton therapy systems.