Research On Vertical Traffic Passenger Flow Statistics Based On Computer Vision

With the continuous development of science and technology,more and more buildings tend to be intelligent,and the comprehensive buildings are emerging.The traffic flow in the building also has complex changes.The traditional elevator scheduling strategy can not meet the demand of fast changing vertical traffic flow.The accurate elevator traffic flow data is the basis of elevator traffic pattern recognition,and is also the data support of the follow-up elevator decision-making,so the research of elevator passenger flow statistical system is of practical significance.But the accuracy of the traditional elevator traffic flow statistical method has shown insufficient.It is very important to propose a more intelligent elevator traffic flow data method,to collect traffic flow data more stable and accurate.In this thesis,from the point of computer vision,the method based on depth feature detection and filter tracking is proposed to collect the passenger flow data.It can collect passenger flow data effectively,and provide a new idea for the collection of elevator traffic flow data.The main work of this thesis is as follows:(1)According to the relevant information at home and abroad,we summarize the status quo and trend of elevator passenger flow statistics and pedestrian detection and tracking at home and abroad.And the program using the detection algorithm based on the depth feature and the improved filtering and tracking algorithm to calculate the passenger flow data of the elevator is developed.(2)Aiming at the influence of noise on the depth image,by analyzing the imaging principle of Kinect and using it to obtain the depth information of elevator passenger flow,the preprocessing method of image defect removal is studied.The joint bilateral filter is used to remove image glitch,the interpolation is used to fill the image hole.(3)Aiming at the defects of the traditional RGB detection,this thesis proposes a human detection algorithm based on depth features,which integrates the features of depth difference gradient and depth similarity of human body through Multiple Kernel Learning,effectively reduces the feature dimension and improves detection speed.The detection window is improved by using depth information clustering,and through the classification of SVM,the method to optimize human target detection is discussed.(4)By analyzing the actual situation of the elevator environment and the actual situation of the passengers entering and leaving the elevator,the tracking method based on NN algorithm and Unscented Kalman Filter is proposed.And we apply the feature matching method to solve the bottleneck of NN algorithm.Experimental results show that this method can accurately track the target state and provide conditions for passenger counting.(5)According to the conditions of the application environment,the double counting line method based on detection and tracking is designed to count the number of people,and the accuracy of the method is verified in the simulation and actual environment.