SBIR Phase II: Epipolar-Plane Imaging for Robot 3D Vision

这个小企业创新研究（SBIR）第二阶段项目的更广泛/商业影响旨在改善机器人与人类的互动。目前，机器人涉及社会的许多领域，包括物流、制造、自主导航、视频通信、复杂机械维护/维修任务的远程监督、战场和灾难中的支持，以及各种培训、教育和干预场景中的互动，包括远程医疗。这项技术可以通过更高质量的3D传感、更高精度的可视化和提高工人的生活质量，在工作场所提供更有效的自动化。该技术解决了被动三维场景测量的精度和可靠性问题。该小企业创新研究（SBIR）第二阶段项目致力于从被动获取的图像数据中获取可靠而精确的场景三维表示，用于导航、抓取、操纵和自主系统在不受限制的三维空间中的其他操作。该技术一直是计算机视觉领域的一个长期挑战，许多努力提供了在特定条件下的适当解决方案，但缺乏跨广泛应用的适用性。其他方法通常会提供不准确的结果，例如，视图中存在重复的结构，薄特征，深度范围大，或者结构与捕获几何形状的各个方面对齐。基于图像间特征的匹配，当前的技术在特征具有相似外观时失败。该技术通过低成本使用过采样消除了该过程的不确定性，使用一组特定的额外视角来取代确定性线性滤波的“匹配”。提高3D场景测量的可靠性和精度将为机器人与世界的互动开辟新的机会。该项目的成功将把潜在的光场技术推进到更广泛的应用领域，在这些领域中，使用人工现实/虚拟现实（AR/VR）或混合现实（如远程协作和远程交互）的人在环操作依赖于准确和响应的可视化和场景建模，减少前庭和本体感觉不匹配的影响，这种不匹配可能导致恶心等破坏性影响。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase II project seeks to improve robotic interactions with the humans. Currently, robots are involved in large sectors of society including logistics, manufacturing, autonomous navigation, video communication, remote supervision of complex mechanical maintenance/repair tasks, support in battlefields and disasters, and interactions in various training, educational, and interventional scenarios including telemedicine. This technology may offer more effective automation in the workplace through higher quality 3D sensing, greater precision visualization and increased worker quality of life. The technology addresses precision and reliability of passive 3D scene measurements. This Small Business Innovation Research (SBIR) Phase II project addresses the acquisition of reliable and precise three-dimensional representations of a scene from passively acquired image data for use in navigation, grasping, manipulation, and other operations of autonomous systems in unrestricted three-dimensional spaces. This technology has been a long-standing challenge in the computer vision field, with many efforts providing adequate solutions under certain conditions, but lacking applicability across a breadth of applications. Other approaches typically deliver inaccurate results where there are, for example, repeated structures in the view, thin features, a large range in depth, or where structures align with aspects of the capture geometry. Based on the matching of features across images, current technologies fail when features have similar appearance. This technology removes the uncertainty of this process through a low-cost use of over-sampling, using a specific set of additional perspectives to replace the “matching” with deterministic linear filtering. Increasing the reliability and precision of 3D scene measurements will open new opportunities for robotic interactions with the world. Success in this project will advance the underlying light-field technology to broader application areas where human-in-the-loop operations using artificial reality/virtual reality (AR/VR) or mixed reality (such as remote collaboration and distance interaction) depend on accurate and responsive visualization and scene modeling, reducing influences of vestibular and proprioceptive mismatch that can cause disruptive effects such as nausea.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.