On the Complexity of Object Detection on Real-world Public Transportation Images for Social Distancing Measurement
On the Complexity of Object Detection on Real-world Public Transportation Images for Social Distancing Measurement
复制标题
DOI：
10.1109/ijcnn55064.2022.9891955
复制
发表时间：
2022-02
期刊：
影响因子：
--
通讯作者：
N. F. N. Aznan-N.-F.-N.-Aznan-3493136;John Brennan;D. Bell;J. Jonczyk;Paul Watson
中科院分区：
文献类型：
--
作者：
N. F. N. Aznan-N.-F.-N.-Aznan-3493136;John Brennan;D. Bell;J. Jonczyk;Paul Watson
公共场所的社交距离已成为帮助减少 COVID-19 大流行影响的一个重要方面。利用机器学习的最新进展，文献中已有许多研究通过在公共场所使用监控摄像头进行物体检测来实现社交距离。然而，迄今为止，还没有针对公共交通上的社交距离测量的研究。公共交通环境面临一些独特的挑战，包括低分辨率图像和物理摄像头位置，可能导致乘客部分遮挡，从而难以执行准确的检测。因此，本文研究了在公共交通上进行准确的社交距离测量的挑战。我们使用从伦敦地铁和公交网络拍摄的真实镜头对几种最先进的对象检测算法进行基准测试。这项工作强调了对当前公共交通车载摄像头的图像进行社交距离测量的复杂性。此外，利用预期乘客行为的领域知识，我们尝试使用各种策略来提高检测质量，并显示出比单独使用普通对象检测的改进。
Social distancing in public spaces has become an essential aspect in helping to reduce the impact of the COVID-19 pandemic. Exploiting recent advances in machine learning, there have been many studies in the literature implementing social distancing via object detection through the use of surveillance cameras in public spaces. However, there has been no study of social distance measurement on public transport to date. The public transport setting has some unique challenges, including low-resolution images and physical camera locations that can lead to the partial occlusion of passengers, making it challenging to perform accurate detection. Thus, this paper investigates the challenges of performing accurate social distance measurements on public transportation. We benchmark several state-of-the-art object detection algorithms using real-world footage taken from the London Underground and bus network. The work highlights the complexity of performing social distancing measurements on images from current public transportation onboard cameras. Further, exploiting domain knowledge of expected passenger behaviour, we attempt to improve the quality of the detections using various strategies and show improvement over using vanilla object detection alone.