Research On The Safe Driving Model Of Intelligent Vehicle Based On Road Linearity

Safety is a major concern for smart vehicles,while reasonable road alignment acts as the foundation and key factor for safe driving.This research focused on the role of road alignment in driving safety,and discussed three typical relevant concerns,i.e.,emergency braking under extreme driving conditions for small radius curves,prediction of potential accident blackspots as a result of inappropriate road alignment,and detection of abnormal driving trajectory.The main innovative works are as follows:This paper studies the performance of emergency braking under extreme conditions by first building a model of emergency braking distance under typical road alignment conditions.The emergency braking distance models for straights and small radius curves were established,respectively,based on the 2-norm theory.Their correlation was indicated by the Pearson linear correlation coefficient.Then,the data of braking force were clustered based on cosine similarity and relative error,and time series breakout detection was used to divide it into four stages and corresponding time nodes: resistance,accumulation,continuation,and release.A microscopic analysis was then carried out from the perspective of the temporal behaviors of each braking force segment.The experimental results show that under the same initial braking speed condition,the emergency braking distance of the limit small radius circular curve section is 1.1～1.3 times the emergency braking distance of the linear section.This paper studies the effect of road alignment on the driving safety of intelligent vehicles by establishing a neural network model and multivariate mathematical model serving to pre-judge accident-prone locations.Taking road alignment specifications as the input,using a combination of data-driven and model-driven methods to extract features and build models for predicting the number of accidents on the road ahead;The partial correlation coefficients of all the road alignment parameters and the number of accidents were calculated,from which,the characteristics of road alignment with higher levels of relevance to the number of accidents were selected.The influences of combinations of such characteristics and single characteristic on the number of accidents were then verified with T-test and F-test.The safety degree was used as an indicator to judge the safety degree of the road section.The safety degree of the road section was calculated according to the actual measured value,the predicted value and the theoretical value of the number of accidents,and the safety degree was used to pre-judge accident-prone locations,and whether the road ahead was accident-prone sections according to the safety degree.The basic pattern of normal vehicle paths and the basic principle of horizontal alignment were studied.Based on curve similarity,composition of road alignment and geometrical parameters as well as the number of two consecutive lane changes,a comprehensive evaluation method of abnormal smart vehicle path was proposed.An abnormal track database was then established to further improve the traditional Fréchet distance algorithm,and the improved Fréchet distance was used as a measure to make a curve similarity comparison between the trajectory detection sequence and the trajectories in the abnormal track database.The five-point orthogonal fitting curvature method and the orthogonal least square method were used to extract the three basic linear elements(straight line,circle,and transition curve)and the corresponding geometric parameters that make up the entire trajectory.Based on the linear trajectory combination and the geometric parameters of the linear elements,it could be judged as to whether there was any abnormality in the driving trajectory of the vehicle.Regarding roads with unidentifiable lanes,the maximum and minimum values of the to-be-detected trajectory sequence ordinate were determined to calculate the number of two consecutive lane changes accordingly.Afterward,whether the vehicle’s driving trajectory was abnormal was judged based on a comprehensive consideration of the said three indicators,i.e.,curve similarity,linear composition,as well as the geometric parameters and the number of two consecutive lane changes.The established models and methods are verified through actual measurement data.In order to further verify the correctness and generalization of the model,relying on the school’s vehicle networking and smart car test field,a comprehensive test platform for the model built in this article was built to conduct comprehensive tests on real vehicles.Considering the flexibility of programming implementation and test scenarios,and based on the automata theory in the principles of computer compilation,a finite automata model of the smart car comprehensive test scenario is established.The results show that:(1)the Pearson linear correlation coefficient is 0.996 between the straight and curve models,indicating their high linear correlation;if the driving speed is the same before braking,the braking distance along small radius curves will be about 1.2-1.3 times of that along straights.(2)both the neural network model and the multivariate mathematical model can correctly predict potential accident black spots;considering the calculation speed and the interpret ability of parameters,the multivariate mathematical model is preferred.(3)the average accuracy of abnormal driving trajectory detection is 89%;as for its similarity detection,the average accuracy of the traditional Fréchet distance algorithm is 67%,while the improved algorithm increases it to 91%.