To investigate the characteristics of electric bicycle overtaking behavior in urban non-motorized lanes, multidimensional kinematic indicators were extracted from unmanned aerial vehicle aerial survey data, and a Log-Logistic Accelerated Failure Time (AFT) model incorporating riders’ social attributes and microscopic riding styles was developed to analyze the overall influencing factors of overtaking duration. Meanwhile, considering the non-negative and right-skewed statistical characteristics of electric bicycle overtaking duration, a Log-Quantile Regression (L-QR) model was introduced to reveal the heterogeneous effects of various influencing factors at different quantiles. The AFT model shows that overtaking distance, the initial and final speeds of the overtaking vehicle, initial speed difference, rider type, and riding style all have significant effects on overtaking duration; for each one-unit increase in overtaking distance, the expected overtaking duration increases by 2.35%; for each one-unit increase in the initial and final speeds of the overtaking vehicle, the overtaking duration decreases by 7.39% and 6.03%, respectively; compared with ordinary riders, food delivery riders have a significantly shorter overtaking duration, with a reduction of 2.37%. The L-QR model further reveals the quantile heterogeneity of the influencing mechanisms, indicating that overtaking distance and vehicle speed have globally significant effects on overtaking? duration, whereas riders’ social attributes and riding styles significantly affect overtaking duration only in the quantile range of 0.25≤τ≤0.75 . In extremely short- and long-duration overtaking scenarios, their statistical significance decreases, and their marginal effects are weakened due to the constraints of the traffic environment. The findings can provide a theoretical basis for optimizing non-motorized lane design and improving traffic management.