通過深度學習評估公共開放空間的利用率：以底特律河岸開放空間研究為例
- 2020 年 2 月 17 日
- 筆記
hi，大家好~我是shadow，一枚設計師/全棧工程師/算法研究員，目前主要研究方向是人工智能寫作和人工智能設計，當然偶爾也會跨界到人工智能藝術及其他各種AI產品。這是我發在《人工智能Mix》的一篇論文閱讀筆記。
文末了解《人工智能Mix》
Measuring the Utilization of Public Open Spaces by Deep Learning: a Benchmark Study at the Detroit Riverfront
通過深度學習評估公共開放空間的利用率：以底特律河岸開放空間研究為例
體育活動和社交活動是確保健康生活方式的基本活動。公園、廣場和綠道等公共開放空間（Public open spaces ，POS）是進行這些活動的關鍵環境。為了評估POS，需要研究人類如何使用POS中的設備。然而，研究POS使用率的傳統方法是手工、耗費時間和人力的，往往也只定性研究。
因此，利用監控攝像機採集的畫面，通過計算機視覺提取用戶的相關信息是一個重要的研究課題。
– POC概念驗證
該論文提出了一個概念驗證，使用深度學習計算機視覺框架（主要是Mask R-CNN），用於定量評估POS中的人類活動，並以底特律河濱保護區（DRFC）為例，對提出的框架進行了實例研究。
– 數據集
本文還構建了一個數據集 「Objects in Public Open Spaces」 (OPOS)，該數據集包含了從18個攝像機採集的7826幅在不同照明條件下穿過DRFC停車場的帶標註的圖像。
– 應用
該框架自動生成行為地圖以定位不同的POS用戶：
原文摘要
Physical activities and social interactions are essential activities that ensure a healthy lifestyle. Public open spaces (POS), such as parks, plazas and greenways, are key en- vironments that encourage those activities. To evaluate a POS, there is a need to study how humans use the facilities within it. However, traditional approaches to studying use of POS are manual and therefore time and labor intensive. They also may only provide qualitative insights.
It is ap- pealing to make use of surveillance cameras and to extract user-related information through computer vision.
This pa- per proposes a proof-of-concept deep learning computer vision framework for measuring human activities quanti- tatively in POS and demonstrates a case study of the pro- posed framework using the Detroit Riverfront Conservancy (DRFC) surveillance camera network.
A custom image dataset is presented to train the framework; the dataset in- cludes 7826 fully annotated images collected from 18 cam- eras across the DRFC park space under various illumina- tion conditions. Dataset analysis is also provided as well as a baseline model for one-step user localization and activ- ity recognition. The mAP results are 77.5% for pedestrian detection and 81.6% for cyclist detection. Behavioral maps are autonomously generated by the framework to locate dif- ferent POS users and the average error for behavioral lo- calization is within 10 cm.
