SBIR Phase I: Scientific Discovery Translation of Snow-Covered Road Perception Software to a Lane Detection in Snow (LDIS) Product

这项小企业创新研究（SBIR）第一阶段项目的更广泛/商业影响是改善汽车运输的安全性、可用性和公众公平性，减少美国每年5300人死亡，418,000人受伤，以及因恶劣天气事故造成的数十亿美元损失。该技术通过摄像头数据处理，在拥堵的十字路口和桥梁、黑暗的隧道、阳光刺眼和降雪等恶劣驾驶条件下识别车道。解决这些问题还可以提高美国在全球汽车市场上的技术竞争力，开发与国防和能源效率应用相关的技术，扩大现有大学课程，并吸引代表性不足的社区参与创业。本研究的基础是利用相机和全球定位数据，专门用于雪地导航，使用实时机器学习方法，而不过度依赖深度学习。这项技术可以在现有的车辆上实施，从而产生广泛的商业影响，并成为发展可行业务的有力手段，从而产生税收，并为当地社区提供技术工作。这项工作的强大技术创新是使用弹性工程方法，单独调整分类，相机和GPS融合以及快速处理机器学习构建的分层计算机视觉系统。该系统在不过度依赖深度学习的情况下，根据人类观察到的地面事实验证成功的性能，因此汽车公司可以使用标准实践成功验证它。这一创新使当前的驾驶辅助产品在最需要的时候保持功能：在低能见度、低牵引力的情况下。本研究旨在验证双车道交叉路口、桥梁、隧道、阳光刺眼条件下、100多英里活跃降雪以及误导性环境信息的创新。将收集这些实例的数据，并对现有技术进行修改和改进。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 I project is improved automotive transportation safety, usability, and equity for the general public which reduces the annual 5,300 fatalities, 418,000 injuries, and billion-dollar losses from inclement weather crashes in the United States. The technology identifies the driving lane using camera data processing in challenging driving conditions such as congested intersections and bridges, dark tunnels, and during sun glare and active snowfall. Addressing these problems also enables U.S. technology competitiveness in the global automotive market, development of technologies relevant to national defense and energy efficiency applications, expansions of existing university courses, and entrepreneurial engagement from underrepresented communities. The foundation for the proposed research is the utilization of camera and global positioning data specifically for navigation in snow using real-time machine learning methods without an overreliance on deep learning. This technology can be implemented in current vehicles, enabling a widespread commercial impact and a strong means to grow a viable business that is generating tax revenue and offering technology jobs to the local community.The strong technical innovation of this work is a hierarchical computer vision system built using a resilience engineering methodology, individually tuned classifications, camera and GPS fusion, and fast processing machine learning. This system provides verification of successful performance with respect to human-observed ground truth without an overreliance on deep learning so that it can be successfully validated by automotive companies using standard practices. This innovation allows current driving assistance products to remain functional when they are needed most: in low visibility, low traction situations. This research aims to verify the innovation in two-lane intersections, bridges, tunnels, under sun glare conditions, in 100+ miles of active snowfall, and in instances of misleading environmental information. Data for these instances will be collected and the existing technology will be modified and improved.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.