Research On Drving Intention Recognition Algorithm For Vehicle Safety Assistant Systems

Nowadays the vehicle safety assistant systems(VSAS) have received much moreattention gradually, with their ability to assist the drivers to control the travel conditionof vehicle, and try to avoid traffic accidents as much as possible. In a closed-loopcontrol system of driver-vehicle-road, VSAS consider the driving intention, the trafficconditions and the travel condition, then develop the appropriate control strategiescontrol the travel condition of vehicle. Therefore, the fisrt step is resognising the drivingintention. In order to recognise the driving intention which is needed to touch off theVSAS effectively, this thesis does the folloing research works:Firstly, research and analyze the current widely used several kinds of VSAS,oriented to5of them, determine the corresponding driving intentions need to berecognise. Research the relationship between the driving intention and drivingbehaviour, then based on the hierarchical and sequential characteristics of HiddenMarkov Model(HMM), a double-layer HMM structure is set up, which is divided intobehaviour-layer and intention-layer to recognise the driving behaviour and drivingintention respectively. Through the analysis of vehicle dynamics model and comparisonof the simulation results, choose the observed sequence parameters for each recognitionmodel.Secondly, use TESIS DYNAware to collect experiment data for each drivingintention, based on the improved Nair test and K-means algorithm to eliminate theabnormal data segments and to set the critical value of eigenvalues for each drivingbehaviour. To recognise the short-term, simple driving behavior, an improved hiddenmarkov model is used, and the behaviour-layer is divided into several moduels.Moreover, after recognising the driving behaviour, analysis the driving behaviourmanifestation of each driving intention, use the recognised driving behaviour as theobserved sequence of intention-layer, then off-line train the models. Optimize theparameters of double-layer HMM through the simulation result comparison, the off-linerecognition rate is improved.At last, to recognise real-time driving intention, a online recognition model is built. Based on the off-line traind parameter, choose the appropriate recognition period andreal-time recognition experiments of the5driving intentions are tested. The simulationresult shows that, MGHMMs in the behaviour-layer can recognise the short-term,simple driving behaviour in real time effectively, MDHMMs in the intention-layer canrecognise the long-term, complex, real-time driving intention well.