Reconstruction of Pedestrian Itineraries in Shopping Streets Using Sparse Image Data

Abstract

The use of computer vision technology to analyze video or image data provides a promising method for collecting individual pedestrian behavior data in shopping streets. This paper is specifically focused on situations where the image data are sparse and only limited information about pedestrians' movements is captured. A method for reconstructing the shopping itineraries of pedestrians is proposed and validated. The method involves using computer vision algorithms to identify specific pedestrians in the images, and estimating the durations of visits to different places in the shopping street. This is achieved through the use of a recursive least squares model. The paper demonstrates, through simulation-based validation, that mean visit durations can be accurately and reliably estimated with a sufficiently large sample size, and the relative performance of the shopping street can be reliably measured. The empirical validation of the method utilizes pedestrian behavior data collected from East Nanjing Road in Shanghai, China, over the past two decades. By comparing the mean visit durations and relative performance of the street over the years, it is found that these longitudinal changes can be explained by the development of retail and spatial improvements on the street; this further supports the proposed method.