SBIR Phase I: Automated Census of Street Trees from Public Imagery

这个小型企业创新研究(SBIR)第一阶段项目的更广泛的影响/商业潜力是，它将在美国各地带来更多和更健康的城市树木。它将影响气候、能源、空气和水的质量以及宜居性。这将通过市政当局更有效和更高效地管理城市森林，以及通过更多和更有组织的公民参与来实现。树木对一个社区来说是一笔宝贵的资产。它们的好处包括：减少能源消耗；改善空气和水的质量；增加碳捕获和储存；增加财产价值；改善个人和社区的福祉。为了确保最大效益和最低成本，行道树需要得到有效的管理。为了做到这一点，市政当局需要最新的树木清单。这个小型企业创新研究(SBIR)第一阶段项目旨在利用尖端的计算机视觉和机器学习技术，根据航拍和街景图像自动生成行道树清单。清单包括每棵树的GPS位置、树种、树干直径和对其健康状况的估计。每当有新的航拍或街景图像可用时，库存就会更新。树木检测和树种分类都是前所未有的应用，它们基于深度卷积网络、3D几何和由专家和众包相结合收集的大型标注数据集的学习能力。这一创新使得能够以当前成本的一小部分来维护始终最新的、准确的、完整的全美行道树库存。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is that it will lead to more numerous and healthier city trees throughout the US. It will impact climate, energy, air and water quality and livability. This will be achieved through more effective and efficient management of the urban forest by municipalities, and through more and better organized citizen participation. Trees are a valuable asset for a community. Their benefits include: a reduction in energy use; improvement in air and water quality; increased carbon capture and storage; increased property values; and an improvement in individual and community well-being. To ensure maximum benefit and minimum cost, street trees need to be managed efficiently. In order to do this, municipalities need up-to-date inventories of trees. This Small Business Innovation Research (SBIR) Phase I project intends to automatically generate inventories of street trees from aerial and street-view imagery using cutting-edge computer vision and machine learning techniques. The inventory contains each tree's GPS location, its species, trunk diameter and an estimate of its health. The inventory is updated every time a new aerial or street-view image becomes available. Both tree detection and tree species classification are unprecedented applications based on combining the learning capabilities of deep convolutional networks, 3D geometry, and large annotated datasets that are collected by a combination of experts and crowdsourcing. This innovation makes it possible to maintain an always up-to-date, accurate, complete, US-wide street tree inventory at a fraction of the current cost.