Research On Technique Of Vehicle Assistant Driving Based On Fusion Of Infrared And Visible Light

In recent years,with the continuous improvement of national income,people trend to buy cars to meet their travel needs.Due to this,traffic accidents happen frequently.Through the analysis of the statistics,there are two main reasons cause traffic accidents:the one is fatigue driving;the other is lack of light or the sight is blocked,such as driving at night or driving in rainy and snowy days.The driver can not make correct operations according to the road environment ahead in those times.More and more automobile manufacturers and institutes focus on the research of automobile assisted driving technology have invented due to these circumstances.Automobile assisted driving technology can reduce traffic accidents effectively.Through the analysis,most of the current vehicle assisted driving technology is mainly based on visual sensor,which has good performance in sunny days,and poor performance in nights.None of the current automobile assisted driving technology can truly realize all day auxiliary driving.However,with the development of infrared imaging technology,the short board of visible light equipment has been supplemented,making it possible for all day automobile driving assistance.Based on the fusion image,this thesis studies each stage of traditional target recognition technology,which improves the accuracy of vehicle assisted driving system for passenger and vehicle recognition.Besides,it also improves the adaptability for complex scenes.In the image preprocessing stage,this thesis studies both images’ principles and characteristics.According to the connections and differences between them,the median filter algorithm is improved in the image denoising stage.In the stage of image fusion,the NSCT transform fusion algorithm is improved.In the stage of target detection,after studied two existing target detection algorithms,an improved algorithm which makes full use of morphology and wavelet transform is proposed.In the stage of target location,according to the analysis of the requirements,this thesis choose the target location algorithm based on the gravity principle finally.Based on the study of three kinds of features,this thesis raise a improved HOG-LBP feature.In the stage of target recognition,this thesis studies SVM classifier and Ada Boost classifier.SVM classifier is used as weak classifier of Ada Boost classifier,which makes Ada Boost classifier have the ability to deal with high-dimensional features.The experiment’s result shows that the improved algorithm can recognize pedestrian and vehicles in the fusion videos effectively.