<p>Accurate identification of pedestrian crossing intention is the key to ensuring pedestrian safety, and the effective analysis of crossing behavior characteristics is one of the core requirements for the establishment of crossing intention recognition model. In this study, pedestrian intentions were classified as waiting for crossing and direct crossing, and the influences of the absence/presence of zebra stripes and the pedestrian age on pedestrian street-crossing intentions were compared based on 3600 effective samples extracted from a real road. The results demonstrate that the two factors have significant effects on the pedestrian crossing percentage, the waiting time, the accepted time to zebra stripes (TTZ), the rejected TTZ, and other characteristic parameters. Hence, to ameliorate the performance of the identification model, an ensemble learning method with a stacking framework is proposed for the identification of pedestrian waiting and direct crossing intentions. The distance between the pedestrian and the zebra stripes, the distance between the vehicle and the zebra stripes, the vehicle velocity, the pedestrian walking speed, the TTZ, the safe vehicle deceleration, waiting time, pedestrian age, and absence/presence of zebra stripes are employed as the inputs of the recognition model. The identification results indicate that the accuracy of the proposed model reaches 96.7% when pedestrians arrive at the road curb, which is the highest accuracy as compared to those of other models. The research conclusions could provide technical support for intelligent driving systems (IDSs) to more accurately recognize pedestrian street-crossing intentions, and can provide a basis for IDSs decision-making and interaction with pedestrian.</p>