Research On Vehicle Detection And Tracking Algorithm In Blind Area

With the rapid development of China's economy,people's living standards improved significantly,car ownership was explosive growth,traffic safety has become a problem cannot be ignored.The advanced driver assistant systems can effectively drive the driver to drive safely through sensors,pattern recognition and artificial intelligence.It has been extensively researched at home and abroad.The blind spot vehicle real-time detection technology as a key content in advanced driver assistance systems,to achieve lane change,collision avoidance and overtaking and other auxiliary driving applications is of great significance.In this paper,a blind spot vehicle detection and tracking algorithm based on monocular vision is researched and implemented in order to solve the existing problems of blind vehicle detection based on visual characteristics.Based on the actual road video test to show better results.The main work of this paper is as follows:(1)Based on CAdaBoost blind vehicle detection algorithmA concurrent adaptive boosting classification algorithm is proposed and applied to the blind vehicle detection,based on the parallel strategy,which is mainly based on the problem that the large number of samples are off-line training time is longer and the real-time and robustness of the blind vehicle detection process is poor during the online detection process.Firstly,the use of parallel training at the same time a number of weak classifier to improve the speed of offline training.Then,the weighting parameters are introduced for each weak classifier weighting coefficient,so that the weight of the weak classifier with the small classification error rate is higher,and the role played by the final weighted voting is bigger.And the weak classifier with a high classification error rate has a lower weight,which makes it less effective in voting,so that the role of each weak classifier in the final strong classifier can be more accurately described to improve the accuracy.Finally,in the training of the classification model,the blind area vehicle real-time detection.(2)Vehicle tracking algorithm based on MCAKCF blind zoneA multi-scale adaptive kernel correlation filtering algorithm racking algorithm is proposed to solve the problem that the tracking performance of the kernel-related filter tracking algorithm is reduced under the target scale and occlusion,when the target scale is changed and the tracking performance is poor.And apply it to blind vehicle tracking.Firstly,a set of sample sets of multiple scales is constructed by scale scaling factor for the input detection target frame.Then,the feature sequence of each sample set is extracted and the response value of each sample set is calculated.On this basis,the position of the current target vehicle is chosen as the current target.Finally,for the next frame image,the validity of the tracking target is judged by defining the peak mean value as the judgment index,and the current model is allowed to be updated only when the peak mean value is greater than the set threshold,when the peak mean value is less than the set threshold will keep the previous frame of the model to be updated.Based on the international standard data set and the actual road data set,the results show that multi-scale adaptive kernelized correlation filters not only shows better results on the target scale change and occlusion problem,but also compared with the correlation filter algorithm and the classical tracking algorithm,and in the integrity of the assessment has good stability and real-time.(3)Blind spot vehicle ranging algorithm based on monocular visionIn order to carry out early warning of blind vehicles,this paper based on the principle of camera aperture imaging,to achieve the use of monocular camera blind vehicle distance measurement.Firstly,use the camera to shoot multiple chess grid images for offline calibration,get the camera's internal parameters and distortion coefficient,at the same time,by measuring the height of the installation.Then,the distortion of the detected blind area vehicle image is corrected by using the obtained distortion parameter.Finally,calculating the actual distance between the detected and the detected plurality of vehicle targets with the calibration of the internal parameters and the actual installation height,and the warning distance is compared with the safety threshold according to the calculated distance The experimental results not only validate the adaptability and effectiveness of the method,but also lay the foundation for the application of blind vehicle detection technology in vehicle aided driving.