Research On Several Key Problems Of Vehicle Lane Change Warning

Vehicle lane change warning system using sensor technology, pattern recognition,artificial intelligence, or other technology means to evaluate the risk of lane change process. Ifthe lane change process was dangerous, this system will interference the driving process and itwas important to ensure the safe running of vehicles. However, some of technical problemsrelated to lane change warning were not been investigation deeply, such as identification andconfirmation of real lane change behavior, forecasting and analyzing of lane change conflict,lane change warning individuation requirement of different manner drivers and so on. Aimingat these problems, several key problems of lane change warning system were investigated inorder to reduce false alarm rate and false negative rate based on real lane change dataanalysis.Radar sensors, lane line identification system, GPS, gyroscope, CAN bus data capturecard and other instruments were used to gather characterization parameters synchronization.Driving manipulation parameters, vehicle running parameters and road environmentparameters were collected.53drivers carried out real road driving in different road conditions,above2000natural lane change data were gathered. Based on this real road test data, severalkey problems of lane change warning system were researched by using patter recognitiontheory, the main research contents were as follows:1. General lane change recognition model and fast lane cross lane change recognitionmodel were established by using real road lane change behavior test data. Adopting Kalmanfiltering to deal with original data can weaken data step evolution which was caused by theinstrument’s precision. For general lane change behavior, by using data normalization, datadimension reduction method, and using genetic algorithm to optimize the parameters of SVMmodel, the classify accuracy rate of training set and the recognize accuracy rate of test setwere improved.2. Lane change line crossing time forecast model was established based on real road lanechange trajectories. Adopting seven polynomial curve model to fitting the real lane changetrajectories and the result shows that the seven polynomial curves had the advantage of flexible curve trend, high fitting accuracy and simple math model, it can be used to fittinglane change trajectories in different speed and steering angle. By analyzing a large number oflane change trajectories, the result shows that similarity was existed between trajectories.Time to line crossing was forecasted based on seven polynomial curve model and trajectorysimilarity, the test result shows that vast majority of forecast time were less than or equal to0.1seconds.3. Position distinguish model among vehicles was established based on road curvatureestimating. Road curvature was estimated based on vehicle yawrate and speed and theestimated accuracy can be improved by data filtering method, but the curvature estimatedresult contained serious noise. In straight sections road and bend sections road, the positiondistinguish model of own vehicle and other vehicles in rear area was established by using theestimated road curvature, relative angle and relative distance of own vehicle and othervehicles, potential dangerous vehicle of lane change process was distinguished by using thismodel. Model verifying test result shows that this model was not sensitive to the measureerrors of road radius, when the radius changed in a larger range, the distinguish result of thismodel keep stable.4. By analyzing real vehicle test data, driver’s manner were classified into aggressivedrivers, relatively aggressive drivers, relatively cautious drivers and cautious drivers in themethod of fuzzy mathematics. In case of over50km/h speed, lane change warning rules fordifferent manner drivers were established based on TTC and relative distance. Verification testresult shows that the fuzzy driving manner classify model can distinguish different typedrivers correctly, and the lane change warning rules based on diving manner can realize lanechange warning function for different drivers.The research was sponsored by National Natural Science Foundation (51178053) andNational Key Technology R&D Program (2009BAG13A05).