Beyond the Frame: Evaluating Panoramic vs. Perspective Images for Assessing Place Perception

About this datasetThis dataset was collected Urban Panoramic Perception (UP2), an online platform to evaluate the use of panoramic imagery in urban perception studies.AbstractStreet View Imagery is used extensively to predict and understand the perception of the urban space by pedestrians (e.g., feeling of safety). While traditional approaches use perspective images (implying a limited field of view) to collect the opinion of participants, this paper evaluates the use of 360° imagery in crowdsourcing experiments. 360° imagery has the benefit of adding visual context compared to perspective images (``out-of-frame” information) and provides a better immersion for the participant, while being accessible on regular computer monitors. We developed a website to collect the opinion of 364 participants on both types of images. Based on 48 comparisons, our experiment shows that in 19\% of the cases ,the evaluation of the safety of a place differs if the context is shown to the participant beforehand. Additionally, we demonstrate that the out-of-frame information plays a role in how a participant rates an image. Finally, we provide comparisons with a Machine Learning model trained on the popular Place Pulse 2.0 dataset and show that using 360° imagery has a smoothing effect over the ratings of a place, compared to perspective images. These results have strong implications for the creation of urban perception datasets via crowdsourcing techniques.FundingThis work was supported by the École Centrale de Nantes and the Région Pays de la Loire (France) through the AiBy4 program.

数据集说明
本数据集采集自城市全景感知（Urban Panoramic Perception，UP2）在线平台，该平台旨在评估全景影像在城市感知研究中的应用。

研究摘要
街景影像被广泛用于预测和解读行人对城市空间的感知（如安全感）。传统研究方法多采用透视影像（视野范围有限）来收集参与者的评价，而本研究则评估了360°全景影像在众包实验中的应用效果。相较于透视影像，360°全景影像可提供额外的视觉上下文信息（即“框外”信息），能为参与者带来更强的沉浸体验，且可通过普通电脑显示器便捷访问。我们搭建了专属实验网站，收集了364名参与者对两类影像的评价数据。基于48组对照实验，本研究发现：19%的场景中，若提前向参与者展示场景上下文，其对某一地点安全感的评价会出现差异。此外，本研究证实框外信息会影响参与者对影像的评分。最后，我们将实验结果与在热门数据集Place Pulse 2.0上训练的机器学习（Machine Learning）模型进行对比，结果显示：相较于透视影像，使用360°全景影像可让地点评分的分布更为平滑。上述研究结果对通过众包技术构建城市感知数据集具有重要指导意义。

资助说明
本研究由法国南特中央理工学院（École Centrale de Nantes）与卢瓦尔河地区大区（Région Pays de la Loire）通过AiBy4项目资助。