Research On The Judgment Criterion And Identification Method Of Unauthorized Pick-up Behavior Based On Traffic Video

With progression of our times and advancement of science and technology,people have relied heavily on road traffic while also increasing their demands on it.In recent years,the application rate of non-local enforcement has been increasing year by year,and various types of road traffic management techniques have been widely applied by many traffic organizations.In addition,there are automatic identification technologies and forensics standards developed for many common offenses and irregularities.As for the unauthorized pick-up behavior that acts as a traffic violation that happens frequently,development of its forensics standards and automatic identification technologies has the broad application prospects in the field of traffic safety.However,there is no relevant research on it.To address these issues,this paper puts forward a reference for forensics standards of unauthorized pick-up behavior,and initially completes the automatic recognition of the unauthorized pick-up behavior based on video surveillance.The main work is as follows:(1)In this paper,an improved Vibe moving object detection algorithm is proposed.In order to solve the problem concerning the ghost image due to delay in updating of initial background in which there is moving object or in which the stationary object moves suddenly,this paper proposes an improved method which is based on motion state changes and changes the background updating probabilities so as to accelerate ghost elimination.Vibe algorithm is combined with the background-attachment differencing to obtain the foreground of the moving /stationary objects,determine the position.Then,the background updating probability is changed,by using the Vibe algorithm to judge whether there is corresponding object within the range and whether the motion status is changed in the video.The algorithm accelerates ghost elimination on the basis of maintaining real-time capability,with good detection effect.(2)An algorithm for automatic identification of authorized pick-up behavior based on the position relationship between people and vehicles is proposed.Combining with the HOG and Haar-like features,the taxi is detected using the Adaboost classification algorithm,which is then combined with the improved Vibe algorithm to obtain the target scope of the stationary vehicle.By determining the position relationship of the pedestrian targets relative to the scope,the possibility of authorized pick-up behavior is determined.The algorithm can recognize authorized pick-up behavior in a more complicated context.(3)By analyzing behavior differences among getting on the vehicle,walking,running,etc.,this paper selects two different feature extraction methods to extract features of human behavior.One is the feature extraction method based on the body shape and extracting the features of major axis,minor axis,major axis change rate,minor axis change rate,ratio of major axis to minor axis,change rate of ratio of major axis to minor axis and angular variation.The other extraction method is based on motion history,in which the Cmotion features are extracted.Upon comparison of feature differences among different behaviors,appropriate 5 features vectors are selected to generate 5D feature vector.Then,the threshold method is adopted to identify the behavior,effectively reducing the false detection rate.