In order to reveal the formation and change of rear-end collision risk between the lead and following vehicles on sharp curves effectively, a typical accident-prone sharp curved segment is selected to collect traffic flow data by vertical aerial photography from unmanned aerial vehicles. Vehicle trajectory information is extracted via Tracker to determine the post-encroachment time (PET) variable of rear-end conflict cross the sharp curve. The results show that there is a spatial clustering feature in rear-end conflicts, which mainly concentrate in the upstream of the entry transition curve and the downstream of the circular and exit transition curves. Four types of rear-end conflicts make up 83.24% of all types of conflicts, and PET decreases both at threshold moment and within conflict risk range with the former is even more so than the latter, resulting in declines of DPET (derivative of PET) values. Also, four indicators as speed of FV, acceleration, difference in speed and acceleration between LV and FV have a significant impact on the DPET change at the threshold moment of rear-end conflicts. The mean value of DPET during the conflict evolution process of T10 (LV decelerates and FV accelerates) is the smallest with the PET sequence declines most sharply, which shows a significantly higher risk than other conflict modes.