Study On Identification Of Dangerous Driving Status For Motor Coach Based On Cascaded Monocular Visual Perception Technology

Among all the social problems,traffic accident has become the the most serious one.The main cause of major traffic accident is passenger vehicle accident especially.Although there are more and more safety and assistant driving equipment were developed in China in recent years,most of their robustness was on low level,thus can’t meet the demand in highway road,terrible weather or interference caused by viewpoint drifting.Therefore,it is important to carry out the research on the high level robustness lateral and longitudinal dangerous driving state identification technology to provide technical support for passenger vehicles active control,which has important theoretical significance and engineering practical value.Considering the problem that leakage recognition,over recognition and weak recognition caused by the hardware constraint of focal length in the existing monocular / binocular vision technology.While facing the ground distance between the target vehicle tail feature points in front of the target,there was huge calculation deviation in the traditional longitudinal dangerous driving state identification algorithm during projection.Through vehicle dynamics control,the potential hazards of leakage recognition,over recognition and weak identification were analyzed.The measurement of vertical high precision distance is realizedin the range of 10 to 100 meters based on full coverage binocular vision and a projection error repair distance measurement algorithm was proposed.Offline training of massive samples were performed to extract effective vehicle contour and texture features.Haar-like features were used as target description methods,and Adaboost algorithm was combined to accomplish target recognition.Combing with the calculation model of vehicle safety distance,a longitudinal classification warning model for passenger vehicles was established.Considering the detection accuracy was reduced under the condition of image degradation caused by fog weather,by using existing vehicle lateral deviation warning algorithm.In this paper,we use modular one-way import link algorithm to recover degraded image,an improved dark channel priority algorithm based on bilateral filter was used for image de fog enhancement module,an improved unsharp masking algorithm based on Gauss filter was used as image contrast enhancement module,then the two enhancement modules were imported in one way,and the luminance of enhanced pictures were adjusted.With the interest area equation of road line quantitatively calculated by single cascaded monocular visual vehicle lateral migration model,a degraded image restoration technology with good fog removal effect and real-time performance was obtained.An improved non local mean denoising algorithm based on high gradient region was adopted to effectively remove the random noise in the image.The optimal threshold segmentation was improved and the edge contour information of lane marking line was optimized by deep mining.Lane mark and lane mark width were selected based on multi feature set,and lane point feature point fitting were applied to calculate the road equation in the image coordinate system and recognizing the road marking line.The vehicle lateral yaw identification model based on spatio-temporal information fusion was established by estimating vehicle operating posture in the world coordinate system and reconstructing of road key information by inverse perspective projection transformation.Considering the problem that the longitudinal warning distortion caused by view driftting can’t be solved in the process of vehicle longitudinal identification,the vehicle transient longitudinal range correction method based on group filtering was applied to study the vehicle longitudinal dynamic warning technology for bumpty road.Front and rear wheel input under the different speed was calculated based on the filtered white noise method,according to the vehicle equipment and visual characteristics of CCD to establish the dynamic mathematical model of CCD visual imaging equipment in different input under transient response of vertical vibration and pitch angle response simulation analysis,a suitable measurement equipment were selected by the simulation results,road test was carried out and the actual road input response of the vehicle CCD visual equipment was obtained.Statistical analysis was used to analysis the feature points of static longitudinal ranging results,CCD initial parameters of pavement response to changes in response to changes in pitch,vertical and the two interaction effect on the longitudinal changes in response to the proposed location were quantitatively analyzed;location algorithm for vehicle longitudinal transient population filtering method based on the view drifting.A high robust traffic vehicle identification system was set up based on the proposed algorithm and the data of road vehicle test were verified.Experimental images and data were verified to test the algorithm of high robust passenger vehicle dangerous driving state identification system.The test results showed that the proposed algorithm was effective and feasible,and the system worked stably and reliably,and achieves the design application requirements.