Study On The Test Method Of Automotive Active Safety Based On Monocular Vision

At present,the number of motor vehicles in China is increasing year by year,and road traffic accidents are in high rate.The safety of vehicles is getting more and more attention.The development of traditional safety of vehicles such as air bags,seat belts and other vehicles has been in a bottleneck.With the development of visual detection technology in recent years,the test method of active safety based on monocular vision is attracting more and more attention.The testing technology of active safety based on monocular vision is combined with computer to test the vehicle in front of the environment,to identify the testing results,automatically identify the dangerous road,help drivers to auxiliary driving safety,as far as possible prevent the road traffic accidents,and to reduce the losses due to traffic accident.But the existing the test of vehicle active safety based on visual recognition rate is low,such as poor robustness and lost tracking problem,which seriously hindered the development and application of vehicle active safety detection based on visuals.Based on the above disadvantages,this paper proposes a vehicle active safety detection method based on monocular vision.The classification of vehicle detection system is presented and in this paper to determine the vehicle location,based on the improved TLD algorithm which could track the vehicle,Finally the car distance detection model is established based on monocular vision,detection of vehicles.The main research contents are as follows:(1)Establish a vehicle grading detection system based on monocular visionThis paper presented the classification of vehicle detection system from coarse to fineaspects,and used the Hog characteristics and support vector machine(SVM)training positive and negative samples,training vehicle rough classifier,and determining the region of interest(ROI)by the classifier to detect vehicles.By improving the Haar-like characteristics and Adaboost algorithm,combining with cascade classifier training,positive and negative cases sample training vehicle classifier,to verifying this ROI through the classifier,the specific location of the vehicle was eventually determine.The vehicle classification detection system established in this paper is used to detect the active safety of vehicles,and proves the effectiveness of this method.(2)Propose a vehicle tracking method based on improved TLD algorithmIn this paper,the advantages and disadvantages of the common vehicle tracking algorithms are compared and the TLD tracking algorithm is improved.Predicting possible rectangular area of the vehicle by Kalman filter,finding out the TLD scan window intersection with the forecast rectangle window,and tracking the window using TLD tracking module helps to realize the vehicle tracking.Finally,the improved TLD algorithm is compared with the former TLD algorithm,and the effectiveness of the improved TLD tracking algorithm is verified.(3)Establishment of vehicle distance detection modelBased on the traditional camera imaging principle,a forward vehicle ranging model based on monocular vision is established.The Open CV function is developed to calibrate the camera,and the calibrated results are brought into the range model,which combines the derivation of the coordinate system to calculate the distance of the vehicle.The research results of this project can not only be used for detection,tracking and ranging of vehicles,but also can be extended to the detection of pedestrians or obstacles in the future.Combined with vehicle assisted driving technology to prepare for unmanned driving,this project has a strong scientific experimental value and a broad prospect on engineering application.