Search

Trajectory reconstruction of pedestrian is of paramount importance to understand crowd dynamics and human movement pattern, which will provide insights to improve building design, facility management and route planning. Camera-based tracking methods have been widely explored with the rapid development of deep learning techniques. When moving to indoor environment, many challenges occur, including occlusions, complex environments and limited camera placement and coverage. Therefore, we propose a novel indoor trajectory reconstruction method using building information modeling (BIM) and graph neural network (GNN). A spatial graph representation is proposed for indoor environment to capture the spatial relationships of indoor areas and monitoring points. Closed circuit television (CCTV) system is integrated with BIM model through camera registration. Pedestrian simulation is conducted based on the BIM model to simulate the pedestrian movement in the considered indoor environment. The simulation results are embedded into the spatial graph for training of GNN. The indoor trajectory reconstruction is implemented as GNN conducts edge classification on the spatial graph

Construction industry has reported among the highest accident and fatality rates over the past decade. In particular, crane lifting is a notably hazardous operation on construction sites, causing fatal accidents like workers being struck by the boom or objects fallen from tower cranes. Manual monitoring by on-site safety officers is labour-intensive and error-prone, while incorporating computer vision techniques into surveillance cameras would enable more automatic and continuous monitoring of construction site operations. However, existing studies for lifting safety mainly detect the presence of individual objects (e.g. workers, crane components), while a methodology is needed to predict their potential collision more proactively before accidents happen. This paper develops a vision-based framework for predictive lifting safety monitoring, including three modules: (1) object detection and classification: targeting at hook and lifting materials to enable danger zone estimation, along with workers and their personal protective equipment; (2) worker movement tracking and prediction: analyzing the historical moving trajectory of each unique worker to foresee his/her future movement in certain period ahead; (3) multi-level safety assessment: issuing predictive warning in real-time upon any crane-worker conflict foreseen. The proposed framework is applicable to real-time site video processing and enables end-to-end lifting safety monitoring with instant alerting upon unsafe scenarios observed. Importantly, the proposed framework predicts the future movement of workers to proactively identify potential site hazard, in order to trigger earlier safety alert for more timely decision-making. With a large video dataset capturing tower crane operations, the proposed framework demonstrates competitive accuracy and computational efficiency in crane-worker conflict prediction, validating its practicality for real-time lifting safety monitoring