Video Based Passenger Flow Detection And Analysis Algorithms And Their Application In Transport Hub

Large transport hub plays an important role in our everyday life. It integrates railway, metro and buses, featuring with complex transit, large flow and large number equipment. How to get rid of risk and maintain efficient operation is an important subject. With the development of technology, video surveillance has been employed more and more into large stations, making big contribution to the safe and efficient operation of stations. Recently, vision based methods are gradually applied in video surveillance. To address the particular situation and requirement of large stations, the research are as following:Firstly, detection and tracking based people counting method is studied. In this article, a detection and tracking based counting method is employed to count passengers walking through aisles and entrance. HOG-PCA feature is employed as the descriptor of people’s head and shoulder, and linear SVM is used as the classifier. Combining HOG and color based histogram, particle filter is employed to track people. The counting method involves detection and tracking, and get rid of repetition by distance and feature similarity. Our framework can work in real time and obtains a high accuracy.Secondly, algorithms on passenger flow estimation and trajectory analysis are studied. To address the situation when it is crowded and heavy occlusion, the framework of feature extraction and regression is used to estimate passenger flow. Global features such as area and gradients as well as local features of HOG are extracted, and as the input of the Gaussian Process Regression model. For flow analysis, topic model is employed to model the far view scene of the station. Combining with the prior knowledge of entrances and exits, unsupervised LDA is employed to cluster the tracklets. After that, clustered trajectories and flow rates are obtained.Thirdly, algorithms on abnormal invasion detection and abandoned objects detection are studied. Based on the definition of abandoned objects, space and time conditions between the owner and luggage are employed to determine whether the luggage is abandoned. We use codebook model to make background subtraction. Foreground objects and background can be discriminated and real-time requirement can be satisfied. Moreover, by constructing the finite state machine, abandoned objects can be effectively detected.Finally, the design and functions of the video surveillance platform are. The video surveillance platform can store and manage videos, and integrate algorithms on passenger flow counting and abandoned objects detection. Moreover, it can analysis on data and display it visually.