A Bi-LSTM approach for modelling movement uncertainty of crowdsourced human trajectories under complex urban environments

Modelling the movement uncertainty of crowdsourced human trajectories in complex urban areas is useful for various human mobility analytics and applications. However, the existing human movement uncertainty modelling approaches only consider the largest movement distance or speed, and fixed sampling and measurement errors, resulting in limited accuracy in uncertainty prediction. To fill this gap, this paper presents a Bi-directional Long Short-Term Memory (Bi-LSTM) assisted framework for modelling the uncertainty of crowdsourced human trajectories under complex urban environments. The proposed movement uncertainty modelling framework adaptively integrates the pedestrian motion detection characteristics, including the real-time gait-length and heading deviation features under detected step period. The characteristics are further combined with the Global Navigation Satellite System (GNSS) originated location, speed and virtual heading information and constructed as the input features for the uncertainty prediction model. Comparison with the existing uncertainty modelling methods is conducted using the real-world datasets, and the results demonstrate the presented Bi-LSTM assisted framework's robust outperformance in achieving more adaptive and accurate movement uncertainty prediction, as measured by multiple metrics. This study provides an accurate and practical solution for modelling the movement uncertainty of human trajectories under complex urban areas, and can support reliable analytics for crowdsourced urban big data.