Counting Method Of The Entrance Pedestrian Flow Based On Depth Images

The entrance people counting is become a hot issue in the current research,which has a widely application.The traditional people counting is generally effected by human occlusion,surrounded,side by side,environmental factors and so on.The advantages of depth images that its properties can effectively overcome the impact of changes in light and environment,make the counting better.The head and shoulder features of depth images avoid the occlusion furthest.But there is a key to solve this effectively statistics is how to correctly detect multiple pedestrian goals as side by side or surrounded.The depth images are regarded as entry point in this thesis.First,for depth image,detecting moving target by improved the mixtures of gaussian model for the whole image.Second,according to the features of head and shoulder to detect maximum stable extreme region of moving targets by using MSER algorithm of multi-threshold segmentation.Finally,pedestrian tracking is achieved by centroid method according to the results of pedestrian detection.And the entrance people flow counting system is also designed and established based on the depth image.The main contents and conclusions of this thesis include:(1)An improved for mixtures of Gaussian model detection of moving targets in depth imagesAccording the method for moving targets detection based on the entrance pedestrian counting,First,extracting the foreground image used the mixtures of Gaussian model.Then,eliminating the error and noise use the threshold method in the moving foreground which has depth information.Finally,improving the accuracy of moving targets detection used the mathematical morphology and median filter.(2)The detection of depth images based on multi-threshold segmentation MSER algorithmIn order to solve the problems which can't be accurately detected using MSER algorithm because of side by side or surrounded resulting in more pedestrian shoulder together in the depth images.In order to improve the accuracy and real-time of pedestrian detection used the MSER algorithm which used multi-threshold segmentation to detect the pedestrian head maximally stable extreme region,and realize the pedestrian detection.At the same time,the contrast experiment of MSER algorithm is done according before and after improvement.(3)The pedestrian target tracking based on pedestrian centroidAccording to the results of pedestrian target detection,the center coordinates of the rectangular frame can be regarded as the centroid coordinates of the pedestrian target.The centroid of the same target in the two adjacent sequence images have the minimum distance similarity,The tracking method based on the centroid of pedestrian target is used,and the tracking effect is compared with that of the kalman filter tracking algorithm.Finally,according to this algorithm to design the entrance people flow counting system,through the experimental data collected in various places of entrance to verify the feasibility of this system,a good solution for the traditional entrance pedestrian flow counting because of the pedestrian occlusion,surrounded,left and right shoulder side by side and other environmental factors caused the counting accurate declining.The results show that the system has high accuracy and well stability.