A Method for Real Time Counting Passersby utilizing Space-time Imagery
Ibrahim Elhossany Elmarhomy Ahmed · Institutional Repositories DataBase (IRDB) · 2014
People counting is a process used to measure the number of people passing by a certain passage, or any marked off area, per unit of time; and it can also measure the direction in which they travel.Manual counting is one of the conventional methods for counting people.People can simply count the number of people passing a confined area by using counter.Although people can count accurately within a short period of time, human labors have limited attention span and reliability when large amount of data has to be analyzed over a long period of time, especially in crowded conditions.Every day, a large number of people move around in all directions in buildings, on roads, railway platforms, and stations, etc. Understanding the number of people and the flow of the persons is useful for efficient promotion of the institution managements and company's sales improvements.On the other hand, most of the deficiencies of manual counting could be handled through automatic people counting systems.In such systems, counting is performed through many approaches among which are a real-time image processing approach.Where a video camera is used to capture video sequences of crossing people and export them to a software package for being processed and interpreted.Counting people and track objects in a video are an important application of computer vision.Computer vision is the science and application of obtaining models, meaning and control information from visual data.It enables artificial intelligence systems to effectively interact with the real world.Counting people approaches using fixed cameras in image processing techniques can be separated into two types.The first one is an overhead camera that counts the number of people crossing a predetermined area.The second is count people based detection and crowd segmentation algorithms.In the overhead camera scenario, many difficulties that arise with traditional side-view surveillance systems are rarely present.Overhead views of crowds are more easily segmented compared with a side-view angle camera that can segment as one continuous object.This thesis proposes a method to automatically count passersby by recording images using virtual, vertical measurement lines.The process of recognizing a passerby is performed using an image sequence obtained from a USB camera placed in a side-view position.While different types of cameras work from three different viewpoints (overhead, front, and side views), the earlier proposed methods were not applicable to the widely installed side-view cameras selected for this work.This new approach uses a side view camera that faced and solved new challenges: (1) two passersby walking in close proximity to each other, at the same time, and in the same direction; (2) two passersby moving simultaneously in opposite directions; (3) a passerby moving in a line followed by another, or more, in quick succession.This thesis introduces an automated method for counting passerby using virtual-vertical measurement lines.The process of recognizing a passerby is carried out using an image sequence obtained from the USB camera.Space-time images are representing the time as a pixel distance which is used to support the algorithm to achieve the accurate counting.The human regions treated using the passerby segmentation process based on the lookup table and labeling.The shape of the human region in space-time images indicates how many people passed by.To handle the problem of mismatching, different color space are used to perform the template matching which chose automatically the best matching.The system using different color spaces to perform the template matching, and automatically select the optimal matching accurately counts passersby with an error rate of approximately 3%, lower than earlier proposed methods.The passersby direction are 100% accuracy determined based on the proposed optimal match.This work uses five characteristics: detecting the position of a person's head, the center of gravity, the human-pixel area, speed of passerby, and the distance between people.These five characteristics enable accurate counting of passersby.The proposed method does not involve optical flow or other algorithms at this level.Instead, human images are extracted and tracked using background subtraction and time-space images.Moreover, a relation between passerby speed and ACKNOWLEDGEMENT First and most of all, I am very grateful to my Creator Almighty ALLAH, the most Merciful and beneficent, for giving me the strength and patience throughout the course of my study at Tokushima University and all my life.My deep thanks go to Professor Kenji Terada, my supervisor, for his guidance, his kind support, continuous encouragement and help throughout the accomplishment of these studies.I appreciate his inexhaustive efforts, unending cooperation and advice, his deep insights that always found solutions when problems supervened and very creative criticism.I will never forget his kindness and patience in dealing with my style and characters, which are very different from Japanese style.I also thank him for introducing me to the image processing fields of Information science and intelligent system.I also wish to thank Dr. Stephen Karungaru, for his guidance, his kind support.My thanks and appreciation goes to all my colleagues