Research On Lane Recognition And Early Warning Based On Machine Vision

With the rapid development of information and artificial intelligence technology,the automobile utilization is constantly increasing and becoming indispensable transportation in people's daily life.In addition to providing convenience,it also brings serious traffic safety issues.It has become urgent to improve the vehicle safety performance to reduce traffic accidents.The research of safety auxiliary driving system has already been investigated.However,lane recognition and warning technology is one of the core technologies in this field and plays a crucial role in the safe driving.The current robustness and accuracy of lane recognition and early warning technology has been low,especially under the circumstance of complex roads.Due to the lack of line,excessive shadow area,low contrast and obstruction of foreign objects,the detection process is mixed with missed detection and false detection,resulting in the reduced detection accuracy,which is difficult to achieve real-time detection and deviation warning of lanes.In this thesis,a new lane recognition and early warning algorithm based on real-time video data is proposed.The contents are as follows:(1)Image preprocessing: Before image processing,the camera is first calibrated to precisely position the target area.After calibration,several preprocessing algorithms are applied for feature extraction and lane fitting.It involves grayscale,median filtering and histogram equalization.As a result,the noise is well inhibited while the region of interest is enhanced greatly,which is beneficial for subsequent feature extraction.(2)Lane detection and identification: Lane detection follows three steps.First,in order to achieve real-time detection,we apply an adaptive segmentation method to extract the interest areas of the road.Then,a series techniques,namely area filling,edge correction and edge extraction,are used to correct the suspected lanes more accurately.Finally,the lane edge can be obtained by using the multi-feature fusion method.Lane identification is divided into three stages.First,Hough transform and Shi-Tomasi corner detection are applied to extract the characteristics of the lanes.Second,dynamic region division mechanism is exploited to accurately classify the feature points on the lane.Third,the improved least square method is used to fit the hyperbolic curve of the lane.The experimental results show that the algorithm has good robustness and real-time performance.(3)Lane deviation warning plays an important role in the safety auxiliary driving system.In this thesis,we use the angle detection method to predict the lane deviation.Through the video stream,we can get the corresponding slope of the lanes and multiple parameters by numerical fitting for each frame.By analyzing a large amount of data,road deviation warning model can be built for lane deviation warning.(4)Performance comparison and design.In this thesis,the source data comes from a camera mounted in the front of a vehicle.The programming is written by c/c++ and OpenCV2.4.9 library.In experiments,4 sets of video data(total of 7200 test frames),including 7 different scenarios,are randomly selected.The test results show that the recognition rate of our algorithm is close to 97.82%,better than other state-of-the-art methods,H-Endpoints,PD-Filter and EKF-Detection.In addition,due to the use of camera calibration,the search range of the image can be narrowed,so that the efficiency of our algorithm can be greatly improved.Moreover,the presented algorithm can detect the lane effectively as well as sound a warning when the lane change or departure happens.