Research On Security Early Warning Method Based On Behavior Analysis Of The Front Vehicles

With the increasingly serious problem of traffic safety,the active safety and driving assistance system draw more and more attention of the public.The driving assistance system,which is based on the information of vehicles,pedestrians and traffic conditions in the road traffic environment,monitors the environment status around the driver in real time,and ensures the safety of the driver.The movement of the vehicle is random and uncertain under the complex road conditions.An effective analysis of the vehicle’s movement intention could better estimate the dangerous situation around the vehicle,identify the potential collision risk of the environment as soon as possible,and thus assist the driver to take action to avoid the dangerous accidents.This thesis proposes a safety warning method based on the behavior analysis of the front vehicles,which mainly includes the detection and tracking of vehicles,the behavior analysis of the vehicle and the risk assessment method.In order to solve the problem of vehicle detection in low cost collision warning system,this thesis proposes a vehicle detection method based on the android smart phone,which predict the movement of the vehicle according to the historical movement,so as to reduce the detection range and improve the detection efficiency.The vehicle detection is realized by the support vector machine.The method can be used to run on the android mobile phone in real time with a frame rate of 9fps,which meets the requirements of the low cost collision warning system.To deal with the problem of multi-lane vehicle detection,a multi-lane vehicle detection method based on road information is proposed and realized on the PC platform.The experimental results show that this method is robust enough to be applied in most weather and illumination conditions.To distinguish the behaviors of the front vehicles,this thesis designs a dual hidden markov method,the steering behavior model and cut-in behavior model are established separately based on the hidden markov method to predict the turn left,straight,turn right and cut-in behaviors of the front vehicle.Firstly,the matching degree of the observation sequence with the two models are calculated by using the forward algorithm,and the maximum of the matching probability is selected to complete the rough classification of the vehicle behaviors,and then the viterbi algorithm is used to identify the specific behaviors,such as turn left or turn right.The concrete content includes the process of model structure selection,model parameter design,data preprocessing,feature extraction,feature encoding,model training,behavior recognition and so on.The experimental results show that the dual hidden markov method could distinguish four kinds of vehicle behaviors with high accuracy.The traditional rules risk assessment method usually has poor tolerance to input data,this thesis proposes a risk assessment method based on the behavior analysis of the front vehicle.Thus,the system can predict the behavior of the front vehicle,greatly improve the risk assessment results,and can adapt to the complex and changeable traffic environment.In addition,according to national standards,this thesis defines two collision warning level,i.e.the preparation of collision warning and collision warning.The auditory and visual cues are used to remind driver.Experimental results show that the method can effectively identify the potential collision risk in the environment,and ensure the safety of the driver.