Research In The Key Technologies Of Lane Departure Warning System Based On Driving Behaviors

LDWS (Lane departure warning system) is one of the most important components of driver assistance system, which assists driver to reduce road traffic accidents caused by lane departure, by means of alarm (vibration, sound and so on). Thus, it is of great practical significance. The root reason of lane departure lies in driver’s failure to follow the lane. However, current LDWS mainly focuses on the lane departure level estimate through lane detection, but overlooks the researches on driving behavior and vehicle response characteristics, resulting in a high rate of error warning and poor adaptability.Beginning with the driving behavior, the logical relationship among driver, vehicle and road environment during lane departure is analyzed thoroughly, and a LDWS based on driving behavior is established. The warning system can acquire information from driver, road environment and vehicle operating parameters by image sensors and vehicle parameter sensors, and the environmental sensing technology based on lane departure is realized. Meanwhile, direction control model is constructed to achieve the predicted evaluation on lane departure. Combing with the evaluation on lane departure and estimate on steering intention, comprehensive evaluation system on lane departure is constructed, which meets the requirement to make the driver keep abreast of the vehicle running status and environmental condition and take appropriate driving measures to ensure safety.Driving environment sensing technology is the key to LDWS. Environment sensing technology of lane departure mainly includes the detection on lane departure, steering intention and response time of steering. Region of interest is proposed based on the assumption of structured road to achieve sub-regional multi-model fitting, in which precision detection of lane markings is achieved. Steering intention is determined according to the structural parameters of vehicle steering and the detection on steering feature of driver’s face. Physiology evaluation is utilitied to determine the correlation between steering response time and attention state, completing the detection on driver’s attention in use of multi-channel information fusion. On the basis of the preview-follow theory, control environment which is based on lane departure is introduced, simplification and modeling of driver behavior are conducted, and a simplified driver model for directional control is proposed. Driver’s front view is selected as signal preception source and the driving view is assumed to be constant. The driving direction model based on two degrees of freedom vehicle model is established. The mathematical model of steering gain is proposed, and direction control model based on LDWS is built. Corresponding control strategy is put forward, and the feasibility of this model is verified in the automotive virtual test environment based on CarSim&Matlab/Simulink co-simulation.Control strategy of comprehensive evaluation criteria is proposed. Parameter identification of steering gain and preview distance is achieved independently by Genetic Algorithms. Furthermore, the influence of different parameters (vehicle speed, steering reaction time and distance preview) on vehicle stability is analyzed.Comprehensive evaluation system of lane departure is constructed, including lane departure evaluation, predicted lane departure evaluation, steering intention evaluation. A SVM-Adaboost strong classifier is built for overall comprehensive evaluation. Lane departure evaluation reflects the current condition of lane departure, predicted lane departure evaluation reflects the trend of lane departure in the next period of time, and steering intention evaluation is used to identify driver’s intention. LDWS containing all three evaluation methods above can effectively reduce the rate of error warning.Finally, the rationality of evalutation system is verified in driving simulator.