Detection and classification of road users for video surveillance applications

该项目的目标是设计新的方法来检测和分类街道和停车场上的道路使用者。通过道路使用者分类，首先将自行车与行人和机动车区分开来，然后将机动车区分为汽车，卡车，公共汽车，救护车，找到解决这个问题的方法将使我们的工业合作伙伴Genetec能够增强其产品的能力。例如，知道车辆的类别允许验证牌照或访问控制设备是否已经被盗或被篡改，或者其可以用于限制某些类型的道路使用者对区域的访问。为了解决这个问题，我们的方法将依靠深度学习技术来学习检测和分类道路使用者。我们将使用一个层次的深度神经网络（DNN）来从道路使用者的粗略分类到更精细的分类，以获得更具体的类别。从这种检测和分类方法的设计中获得的知识将转移到我们的工业合作伙伴Genetec。这将使Genetec在快速发展的视频监控市场中保持竞争力。能够对道路使用者进行分类将使Genetec能够开发更好的工具来监控交通并提高其产品的功能。此外，Prs Bilodeau和Saunier都在致力于计算机视觉在交通领域的应用，特别是交通数据收集和道路安全分析。道路使用者分类尚未在科学文献中广泛讨论。这项工作也将有利于计算机视觉领域的更细粒度的目标检测任务。最后，该项目将培养一名机器学习和数据科学领域的博士生，他们的技能在加拿大工业中需求很高。

The goal of this project is to design new methods to detect and classify road users on streets and parking lots.By classification of road users, it is meant to distinguish first bikes from pedestrians and motorized vehicles,and then to distinguish motorized vehicles into subcategories like cars, trucks, buses, ambulances, etc. Findinga solution to this problem will allow our industrial partner Genetec to enhance the capabilities of their products.For instance, knowing the category of a vehicle allows verifying if a license plate or an access control devicehas been stolen or tampered with, or it can be used to restrict access to an area to certain types of road users. Tosolve this problem, our approach will rely on deep learning techniques to learn to detect and classify road users.We will use a hierarchy of Deep Neural Networks (DNNs) to go from a coarse categorization of road users to afiner one to obtain more specific categories. The knowledge gained from the design of this detection andclassification method will be transferred to our industrial partner Genetec. This will allow Genetec to remaincompetitive in the fast evolving video surveillance market. Being able to categorize road users will allowGenetec to develop better tools to monitor traffic and advance the capabilities of their products. As well, bothPrs Bilodeau and Saunier are working on applications of computer vision to transportation and in particular fortraffic data collection and road safety analysis. Road user classification has not been addressed extensively inthe scientific literature. This work will also benefit the computer vision field for finer grain object detectiontasks. Finally, this project will train one PhD student in the machine learning and data science fields with skillsthat are in high demand in the Canadian industry.