Korean Journal of Computational Design and Engineering

This research proposes a novel methodology for proactively predicting the risk of collisions between workers in industrial settings. While existing systems mainly utilize distance or speed information, this research focuses on predicting future paths of workers based on time series data and quantifying path uncertainty to assess the probability of collisions. To this end, we designed an autoregressive model that combines a temporal attention mechanism with a bidirectional LSTM to learn complex time series patterns and quantitatively evaluate collision probabilities using IoUs. Experimental results show that the proposed method outperforms GRU, FFN, LSTM, and 1D CNN models combined with the LUBE method in terms of path prediction accuracy and collision evaluation reliability. However, the generalizability is limited by using specific environmental data, and future validation in various environments and extensions considering multi-object interaction are needed. This research is expected to contribute to the development of safety management and accident prevention systems in industrial sites.

키워드 Collision Avoidance, Collision Prediction, Risk Assessment, Trajectory Prediction, Trajectory Uncertainty