利用電腦視覺與訊號處理技術於公開直播攝影機進行行人步頻探測

Abstract

在都市規劃、公共安全與經濟發展領域中，行人動態分析提供了重要的見解。行人步頻可做為評估城市行人流動性的一個指標，不僅顯示了城市規劃的效果，亦反映出城市生活節奏（pace of life）。本研究旨在運用電腦視覺與訊號處理技術，進行行人步頻探測，探索行人步頻與城市環境之間的關聯性。 本研究方法包含三個模組：即時影像收集、行人特徵擷取、行人步頻估計。即時影像收集模組從多個公共直播攝影機自動化獲取影像數據；行人特徵擷取模組透過You Only Look Once version 7（YOLOv7）模型與Simple Online and Realtime Tracking（SORT）演算法處理影像數據，以此獲得行人特徵時間序列數據；步頻估計模組則運用訊號處理技術分析這些行人特徵時間序列數據，從而估計出行人步頻。這種非侵入式的分析方法允許在不干擾行人自然行為的情況下進行數據收集與分析，從而提高了數據的真實性和可靠性。 本研究應用此方法於多個公開直播攝影機，以估計和分析行人步頻，並分析不同地理位置和地點類型（居住教育、旅遊、商業、交通）之間的行人步頻差異。透過對這多個影像源的行人步頻累積分布函數（cumulative distribution function, CDF）數據進行K-均值聚類（K-Means clustering）分析，探討這些聚類形成的背後成因。此外，本研究還探討了台灣不同城市之間，收入與行人步頻的相關性。總結而言，本研究為城市規劃者和政策制定者提供了見解，旨在促進智慧城市的發展，並提高城市規劃的效率和效能。

In the fields of urban planning, public safety, and economic development, pedestrian dynamic analysis provides critical insights. Pedestrian cadence serves as an indicator for assessing urban pedestrian mobility, reflecting not only the effectiveness of urban planning but also the pace of life. This study employs computer vision and signal processing technologies to detect pedestrian cadence, aiming to explore the relationship between pedestrian cadence and urban environments. The methodology of this study comprises three modules: Real-Time Image Collection, Pedestrian Feature Extraction, and Pedestrian Cadence Estimation. The Real-Time Image Collection Module automatically captures video data from multiple public live cameras; the Pedestrian Feature Extraction Module processes this data using the You Only Look Once version 7 (YOLOv7) model and Simple Online and Realtime Tracking (SORT) algorithm to acquire time series data of pedestrian features; the Pedestrian Cadence Estimation Module then analyzes these time series data using signal processing techniques to compute pedestrian cadence. This non-invasive analytical approach allows for data collection and analysis without disturbing natural pedestrian behaviors, thereby enhancing the authenticity and reliability of the data. The method of this study has been applied to several public live cameras to estimate and analyze pedestrian cadence, examining the variations in cadence across different geographical locations and types of locations (living and learning, tourism, business, and traffic). Through K-Means clustering analysis of the cumulative distribution function (CDF) data of pedestrian cadence from these locations, analyzing the reasons behind these cluster formations. Additionally, this study explored the correlation between income and pedestrian cadence across different cities in Taiwan. In sum, this study provides insights for urban planners and policymakers, aiming to facilitate the development of smart cities and improve the efficiency and effectiveness of urban planning.