<p>Safe street crossing relies on effective pedestrian-vehicle interaction; a delay in drivers’ recognition of pedestrian intentions can critically increase the risk of collision. Nevertheless, existing research lacks a contrasting exploration of driving behaviour at unsignalized streets for real pedestrians and pedestrian dummies. In this study, we develop a pedestrian crossing intention judgment system to quantify subjective driver assessments of pedestrian crossing intentions, focusing on how drivers adjust their behaviour to real pedestrians and pedestrian dummies. Specifically, we establish a naturalistic driving and pedestrian dummy experimental platform based on the campus unsignalized streets. Then, six different experimental scenarios of pedestrians crossing the street are designed. In addition, a subjective judgment of pedestrians’ crossing intentions (<i>SJPCI</i>) ranging from 0% to 100% is introduced to quantify drivers’ assessments of pedestrian crossing intentions. The proposed system is validated using 123 effective interactions extracted from 1923 natural driving events. Meanwhile, we designed the same dummy experiment based on the 123-group naturalistic driving scenarios. The results show that naturalistic driving experiments have relatively higher <i>SJPCI</i> values, gentler acceleration, and longer time-to-collision (<i>TTC</i>) compared with pedestrian dummy experiments. Effective communication with pedestrians allows drivers to better judge crossing intentions. Our finding is that real crossing pedestrians are significantly less able to have ambiguous intentions, which enhances understandability for drivers.</p>