Aggressive Lane-change Analysis Closing to Intersection Based on UAV Video and Deep Learning

Lane-change is stressful operation for driver, especially when closing to the complex intersection. Accurate analysis the factor that relate to aggressive lane-change is foundation of any future driving safety research. Tracker was using for analyzing aggressive lane-changing behavior that based on amount of naturalistic on-road experiment using an UAV. Over two thousand lane-change video pieces have been recorded, then the selected features are extracted in to matrix. Also, the anonymous drivers and vehicle information is obtained according to license plates. Based on an advanced DBN deep learning algorithm, a relation model of lane-change behavior is constructed by considering the potential influence factors, including vehicle information, driver information, and the driving behavior. Finally, the accuracy of proposed model is validated by comparing to other common models, which approves our model's superior accuracy, feasibility and concreteness.