The use of visual cues to determine the intent of cyclists in traffic

The purpose of this research was to answer the following central questions: 1) How accurate are human observers at predicting the behavior of cyclists as the cyclists approached a crossing? 2) If the accuracy is reliably better than chance, what cues were used to make the predictions? 3) At what distance from the crossing did the most critical cues occur? 4) Can the cues be used in a model that can reliably predict cyclist intent? We present results that show a number of indicators that can be used in to predict the intention of a cyclist, i.e., future actions of a cyclist, e.g., “left turn” or “continue forward” etc. Results of empirical studies show that humans are reasonably good at this type of prediction for a majority of the situations studied. However, some situations seem to contain conflicting information. The results also suggested that human prediction of intention is to a large extent relying on a single “strong” indicator, e.g., that the cyclist makes a clear “head movement”. Several “weaker” indicators that together could be a strong “combined indicator”, or equivalently strong evidence, is likely to be missed or too complex to be handled by humans in real-time. We suggest this line of research can be used to create decision support systems that predict the behavior of cyclists in traffic.