Responsive and Robust Object Detection for Industrial Point Cloud Applications

适用于工业点云应用的响应灵敏、鲁棒的物体检测

• 批准号: 567583-2021
• 负责人: Najjaran, Homayoun
• 金额: $ 1.46万
• 依托单位: University of Victoria
• 依托单位国家: 加拿大
• 项目类别: Alliance Grants
• 财政年份: 2021
• 资助国家: 加拿大
• 起止时间: 2021-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=722623
• 关键词: Responsive; Robust; Object; Detection; Industrial

Our research objective is to create practical three dimensional or shape-based object detection methods that can support high precision industrial applications such as metrology and visual quality inspection. Automated object detection has been a long-standing topic in both academic research and industrial applications. Although end-to-end machine learning algorithms can adaptively be used to detect the objects in 3D point clouds, it requires a large amount of training dataset to perform well. Also, due to algorithmic complexity, it can be too slow for real-time applications. At the same time, classic statistical computer vision methods are proved to be less computationally expensive and have satisfying performance even with a small amount of dataset. However, these methods are not responsive and adaptive in most cases. To achieve the required accuracies and computation performance, it is required that a combination of both the state-of-the-art machine learning methods and the classical statistical methods with their respective advantages are incorporated into a set of software solutions that can be optimally deal with different application scenarios. In this hybrid solution, the problem with the mentioned shortcomings of both methods can be solved. Instead of attempting to maximize detection capability, we aim at finding a balance among algorithmic complexity, robustness and accuracy by combining the aforementioned methods into a practical industrial solution to help higher-level point cloud exploration and decision making. The deployable software tools at the end of this project will impact the Canadian manufacturing sector and beyond, including practical uses in factories, warehouses, and surveillance systems across Canada. Canadian industry will also benefit from the robust object recognition technology developed for the system, which can be readily applied to the growing market of robotic and autonomous vehicle industry. Training of highly qualified personnel for such industries is another important outcome of this research.

我们的研究目标是创造实用的三维或基于形状的物体检测方法，可以支持高精度的工业应用，如计量和视觉质量检测。在学术研究和工业应用中，自动目标检测一直是一个长期存在的话题。虽然端到端机器学习算法可以自适应地检测三维点云中的物体，但它需要大量的训练数据集才能表现良好。此外，由于算法的复杂性，它对于实时应用程序来说可能太慢了。与此同时，经典的统计计算机视觉方法被证明计算成本更低，即使在少量数据集上也有令人满意的性能。然而，在大多数情况下，这些方法没有响应性和适应性。为了达到所需的精度和计算性能，需要将最先进的机器学习方法和经典的统计方法结合起来，各具优势，形成一套能够最优地处理不同应用场景的软件解决方案。在这种混合解决方案中，可以解决两种方法的缺点。我们的目标不是试图最大化检测能力，而是通过将上述方法结合到实际的工业解决方案中，在算法复杂性，鲁棒性和准确性之间找到平衡，以帮助更高层次的点云探索和决策。在该项目结束时，可部署的软件工具将影响加拿大制造业及其他领域，包括在加拿大各地的工厂、仓库和监控系统中的实际应用。加拿大工业也将受益于为该系统开发的强大的目标识别技术，该技术可以很容易地应用于机器人和自动驾驶汽车行业不断增长的市场。为这些行业培训高素质人才是这项研究的另一个重要成果。