Research On The Methods Of Moving Vehicle Shadows Detection And License Plate Shadows Removal In Traffic Scenes

As an important method of the increasingly developing modern urban transport management,Intelligent Traffic has been a great concern,and it has also became the future direction of modern urban transport as the Internet.The vehicle tracking techniques directly affects the judgment,analysis and process of the Intelligent Traffic System,which is one of the key technologies in Intelligent Traffic System.In the traffic videos,shadows are widespread.Regardless of the daytime sunlight or the night lighting from devices will generate shadows.There are various functions of different the shadows in practical application.For example,sometimes the shadows may be a kind of useful information which support the vehicle detection algorithm,but in most cases the shadows are considered to be a distraction from vehicle tracking,segmentation and information statistics.Shadows usually interfere with the vehicle target geometry parameter extraction and measurement.How to effectively detect and remove the traffic scenes shadows has became a hot and difficult problem in the field of Intelligent Traffic.In this paper,the shadow problems in the traffic scenes are introduced in advance.Then,the digital image features of shadows are analyzed,and the existing shadow detection and removal methods are summarized.On the basis of them,this paper researches the detection of vehicle moving shadows and the removal of license plate shadows.The main innovations are in the following two aspects:1.A moving vehicle shadow detection and filtering method based on Zero-tree Wavelet(ZW)mask are proposed in this paper.Firstly,the proposed method converts the foreground containing noise to HSV color space,then take multilevel down-sampling wavelet transform on the S channel and the V channel.Secondly,put the coefficients in different scale sub-bands associated by constructing the ZW mask of the motion foreground,which makes the mask values of fine scale sub-band can get guidance and correction from the father sub-band coefficients,to improve the accuracy of the sub-bands' adaptive threshold value.Further by combining the shadow color characteristics,improve the distinction degree of judging vehicles and shadows.The proposed method solves the problem that the traditional methods are prone to misjudge the strong shadow,vehicles,background,and improves the distinction degree of vehicles and shadows.2.In traffic scenes,license plate shadows likely to have an impact on the LPR algorithm,and lead to increased error rates of the LPR algorithm.For license plate shadow problems in static single images,this paper presents a license plate shadow removal method whichcombines whit K-means clustering and RGB color compensation.Firstly,the method normalizes the RGB color of plates.Secondly,the binary gray plate problem is transformed into a classification problem within the multidimensional space.The method effectively solves the problem that the traditional global threshold method is difficult to correctly distinguish between characters and shadows in plates,and the problem that the local threshold block based method have blockiness and remaining noise.A large number of simulation experiments verify the effectiveness of the proposed two methods.Among these,the motion vehicle shadow detection and filtering method based on Zero-tree Wavelet(ZW)mask can distinguish vehicles and shadows effectively,and have a good shadow detection and discrimination rate.The license plate shadow removal method makes the foundation for the follow-up algorithms of LPR to recognize shadow plates,which can effectively solves the license plate shadow problem in practical applications.