This paper presents a data fusion algorithm which is able to robustly estimate the Time-to-Lane-Crossing (TLC) of a vehicle traveling along a lane on the basis of road images, collected by an on-board videocamera, and kinematic data coming from car sensors. This algorithm is instrumental to built Lane Departure Warning Systems (LDWS) with enhanced predictive capabilities which allow the generation of earlier warnings able to better prevent dangerous driving situations coming from unintentional vehicle lane crossing occurrences. Comparisons with no predictive strategies are carried out and discussed in order to verify the effectiveness of the proposed approach in some critical driving scenarios simulated within the Carsim simulation framework.