With the increasing popularity of active modes (mainly pedestrians and bicycles), the spaces shared amongst these modes are also rising. Such spaces are often referred to as shared spaces and can often be seen in public areas near train stations, shopping streets and educational institutions. Such spaces are said to enhance safety and resolve spatial limitations of the space while the users are seen to exhibit interesting and complex behaviours as they freely interact with different modes approaching from various directions. This creates a need to collect data and understand the behaviour of people within such spaces.
Many studies focusing on shared space interactions are limited by the current methods of data collection and data extraction. Other studies on people movement also face similar issues. Current data extraction approaches using cameras include repetitive and labour-intensive tasks which makes them costly and inefficient. Thus, this study focuses on automating the data extraction approach to obtain the ground-plane trajectories of people using the data recorded from a 3D-stereo vision camera. The data processing framework was divided into three main stages, (i) agent detection, (ii) ground-plane representation and (iii) agent tracking. The agent detection uses a neural-network-based detection model to detect people on the visual video which is then paired with the depth data to represent people on the ground plane. Each person is represented as a single point on the ground plane for every frame of the video. These points were then inputted into a tracking model to provides trajectories of people on the ground plane. Overall, this research integrates the recent state-of-the-art technologies into a single framework to automate the data extraction process and tests this framework in a real-world setting (behind the Amsterdam central station’s shared space area). For each stage, the advantages, challenge and recommendation for further improvements have been made. By automating the data processing of real-world datasets (including pedestrians and cyclists), This study reduces the cost and provides an easy way to collect and process data on real-world movements of pedestrians and cyclists. This will encourage urban planners to better understand people’s movement within their spaces and facilitate data-driven design approaches in the future.