Research On Detection Method Of Backside Target And Lane Change Early-warning System Based On Machine Vision

The survey found that many accidents occurred in the process of lane change.Drivers should not only pay attention to the safe distance with the front vehicle,but also to observe the traffic environment backside,and they have to chnage lane at the same time.A slight negligence can cause traffic accidents.Therefore,the development of a lane change early-warning system is very important to reduce the driver's lane change pressure,to ensure the safety of lane change and improve road capacity.In this paper,the lane change early-warning system is designed based on the backeside traffic environment.The system obtains the rear side image through the visual sensor,extracts the vehicle speed information from the GPS module,and obtains the lateral acceleration information from the acceleration sensing module.At the same time in VS2010 platform,the algrithems are programmed by using OpenCV2.4.10 function.The main work of this paper is shown below.(1)Lane lines in the picture are detected based on weight clustering method.A training sample library of vehicle front face of different angles is established train classifier used in the system.Based on the Haar-llike feature and the training sample library the cascade Adaboost classifier is trained to detect backside car.Then the vehicle horizontal texture information is used to remove the false area accepted by the classifier.After that through the information of the vehicle bottom,the accurate location of the vehicle canbe obtained.(2)Based on the direction angle model,the lateral angle and pitch angle of the camera can be calculated from the coordinate position of the road disappearance point.And then according to the road disappearance point and the backside vehicle,lane line coordinates,relative distance can be detected through the relative position ranging model.At the same time relative velocity is estimated by the basic velocity model.(3)The cross line time is estimated by the sine function lane model which is in order to identify the intention of lane changing.The safe distance model and lane changing warning rules are established according to the threshold of conflict time.Finally,a simulation of lane changing warning rules is made through the model in Matlab.The results show that in terms of real-time and accuracy,the system can meet the requirements of the lane changing warning situation.(4)In verification experiments,the results show that lane line and vehicle detection algorithms are accurate and reliable.In addition,compared with the traditional detection method,the method which based on Haar-like Adaboost classifier and result re-processing can maintain a high level of recall and precision.Through static distance detection test and dynamic distance test,the resaults show that the longitudinal distance error can be controlled at about 5%;lateral distance error can be controlled within 0.2m.The relative velocity test results show that the velocity estimation method is reliability within the detection range of 60 m.The lane change experiments showed that the early-warning system can meet the lane change requirements of real-time property and accuracy.