DETECTION OF HEAD POSITION USING CHAIN CODE ALGORITHM NORFAIZA BINTI FUAD MASTER OF SCIENCE UNIVERSITI PUTRA MALAYSIA

Transcription

1

2 ' *

3 DETECTION OF HEAD POSITION USING CHAIN CODE ALGORITHM NORFAIZA BINTI FUAD MASTER OF SCIENCE UNIVERSITI PUTRA MALAYSIA 2007

4 DETECTION OF HEAD POSITION USING CHAIN CODE ALGORITHM By NORFAIZA BINTI FUAD Thesis Submitted to the School of Graduate Studies, Universiti Putra Malaysia, in Fulfilment of the Requirements for the Degree of Master of Science February 2007

5 DEDICATION To my mother, my father, my brothers, my sister, my father and mother-in-law, my lecturers, my friends and my husband you are the rhythm in my tune, you are the sun and my moon, you are the beach and my wave, you are the glove and I am the hand, you are the station and I am the train, you are the teacher and I am the pupil, you are the suture to my wound, you are the magnet to my pole, you are the sum to my equations and you are the answer to my question. J dedicate this thesis to you. 11

6 Abstract of thesis presented to the Senate of Universiti Putra Malaysia in fulfilment of the requirement for the degree of Master of Science DETECTION OF HEAD POSITION USING CHAIN CODE ALGORITHM By NORFAIZA BINTI FUAD February 2007 Chairman Faculty : Associate Professor Adznan bin Jantan, PhD : Engineering Nowadays, autonomous vision-based system has been applied to handle human job that reliability and efficiency of any intelligent system gain improvement and enhancement. Human head detection is the first step of an autonomous human recognition system. This thesis focuses on a method to recognize and detect a human at the surveillance or highlighted area boundary based their head beside; a simulation system of head detection was developed using image processing. The main contribution of this thesis is it contributes an algorithm of head recognition and detecting which based on image segmentation, Prewitt edge detection and Chain Code algorithm. Static or still images are used as input data for simulation process. The use of Median Filter (MF) method for preprocessing stage is studied and implemented to make low noise for good signification in an image. Prewitt edge detecting (PED) has been applied to present boundary of features in the images in early stage. The image converted to binary image using Threshold Coding (TC) for difference between boundary and background. Features from the image are train using the Chain Code Algorithm (CCA) to do recognition of the crux human head then do detecting process. The low of complexity in mathematical equation is the factor of chosen this method, compares other techniques. Two environments have iii

7 been applied to demonstrate the performance of the system, a person or more was detected for texture background and untextured background. The anahsis. design and development of simulation system are done in Visual C^. All the methods have been tested on image data and the experimental results have demonstrated a robust system. IV

8 Abstrak tesis yang dikemukakan kepada Senat Universiti Putra Malaysia sebagai memenuhi keperluan untuk Ijazah Master Sains PENGESANAN KEDUDUKAN KEPALA MENGGUNAKAN ALGORITMA 'CHAIN CODE' Oleh NORFAIZA BINTI FUAD February 2007 Pengerusi Fakulti : Profesor Madya Adznan bin Jantan, PhD : Kejuruteraan Pada masa kini, sistem autonomi telah diaplikasikan untuk menggantikan tugas manusia supaya kebolehpercayaan dan keberkesanan pada sebarang sistem yang cerdik mencapai pembaikan dan peningkatan. Walaubagaimanapun, pengesanan kepala manusia merupakan iangkah asas bagi sistem pengesanan manusia secara autonomi. Tesis ini tertumpu kepada kaedah pengesanan atau menjejaki manusia di sesuatu kawasan yang diawasi atau di senarai hitam berdasarkan kepada kepala mereka, dimana satu simulasi pengesanan kedudukan kepala menggunakan proses imej telah dibangunkan. Sumbangan utama yang diberikan adalah algoritma pengesanan kepala manusia merujuk kepada pengsegmenan imej, pengesanan tepian Prewitl dan algoritma Chain Code. Gambar pegun digunakan sebagai data input untuk proses simulasi. Penggunanaan kaedah penyaringan tengah (MF) telah dipelajari dan digunakan untuk mengurangkan hingar bagi memberikan kesan nyata yang baik ada imej. Pengesanan tepian Prewitt (PED) telah digunakan bagi mempersembahkan sempadan sifat atau ciri-ciri pada imej diawal peringkat. Imej tersebut di ubah kepada imej binari menggunakan kod Threshold (TC) bagi membezakan sempadan dan latar belakang. Sifat atau ciri-ciri pada imej dilatih oleh v

9 algoritma Chain Code (CCA) untuk mengenal pasti lengkok kepala manusia dan melakukan proses pengesanan. Tahap kesukaran persamaan matematik yang mudah telah menjadi faktor pemilihan teknik ini berbanding teknik-teknik lain. Dua situasi atau keadaan telah dipilih untuk mendemostrasi pencapaian sistem, seorang manusia atau lebih telah di kesan pada latar belakang bercorak dan latar belakang tidak bercorak. Penganalisasian, rekabentuk dan pembangunan sistem simulasi telah menggunakan perisian Visual C++. Semua kaedah telah diuji pada data imej dan keputusan eksperiment membuktikan sistem ini adalah tegap. IV

10 ACKNOWLEDGEMENTS First of all, Syukur Alhamdulillah. Thanks to Allah for blessing me with healthiness, strength and guidance in completing this thesis. I would like to sincerely thank my supervisor Associate Professor Dr. Adznan Bin Jantan, for having pointed me to the right direction, for his enthusiastic and energetic guidance throughout my study and his support, without which, this thesis would not be possible. His endless enthusiasm settings and patience is something to be admired and sought after in both academic settings as well as in life. To my committee member. Dr. Khairi Bin Yusof; I thank him enormously for giving me additional knowledge about image processing during the class, for being such a nice and helpful person and for looking after my work and making valid suggestions. His suggestions have helped making this work more focused. To my beloved husband, Mohd Erwandi bin Marwan who gave his full support to me and helped developing portion of the system. My deepest and utmost gratitude to my family; they have stood by me and witnessed my study and career evolves for the better. They have also been my definite source of constant support. I love you all very much. vii

11 1 certify that an Examination Committee has met on 9 February 2007 to conduct the final examination of Norfaiza binti Fuad on her Master of Science thesis entitled "Detection of Head Position Using Chain Code Algorithm" in accordance with Universiti Pertanian Malaysia (Higher Degree) Act 1980 and Universiti Pertanian Malaysia (Higher Degree) Regulations The Committee recommends that the candidate be awarded the relevant degree. Members of the Examination Committee are as follows: Abd Rahman bin Ramli, PhD Associate Professor Faculty of Engineering Universiti Putra Malaysia (Chairman) Mohd Fadlee bin A. Rasid,, PhD Lecturer Faculty of Engineering Universiti Putra Malaysia (Internal Examiner) Sabira Khatun, PhD Associate Professor Faculty of Engineering Universiti Putra Malaysia (Internal Examiner) Khairuddin bin Omar, PhD Associate Professor Faculty of Technology and Information System Universiti Kebangsaan Malaysia (External Examiner) HASAN D. GHAZALI, PhD Professo: _, Dean School of Graduate Studies Universiti Putra Malaysia Date: 17 MAY 2007 Vlll

12 This thesis submitted to the Senate of University Putra Malaysia and has been accepted as fulfilment of the requirement for the degree of Master of Science. The members of the Supervisory Committee are as follows: Adznan bin Jantan, PhD Associate Professor Faculty of Engineering Universiti Putra Malaysia (Chairman) Khairi bin Yusuf, PhD Lecturer Faculty of Engineering Universiti Putra Malaysia (Member) AINIIDERIS, PhD Professor/Dean School of Graduate Studies University Putra Malaysia Date: 14 JUNE 2007 IV

13 DECLARATION I hereby declare that the thesis is based on my original work except for quotations and citations which have been duly acknowledged. I also declare that it has not been previously or concurrently submitted for any other degree at UPM or other institutions. Date: 10 MARCH 2007 x

14 TABLE OF CONTENTS DEDICATION ABSTARCT ABSTRAK ACKNOWLEDGEMENTS APPROVAL DECLARATION LIST OF TABLES LIST OF FIGURES LIST OF ABBREVIATIONS Page ii iii v vii viii x xiv xv xviii CHAPTER 1 INTRODUCTION 1.1 Introduction Problem Statement Goal Objectives Research Scopes Thesis Layout Summary 7 2 LITERATURE REVIEW 2.1 Input Detection for Vision-based System Grayscale Images True Colour or RGB Images Binary Images Input Requirement for Related Research Human Head Model Head Crux Detection Using Image Processing and Analysis Process PreProcessing Feature Extraction Region based Shape Representation Contour based Shape Representation Estimation and Detection Recognition Related Research Review Comparison of Head Detection Discussion Summary 37 3 MATERIALS AND METHODS 3.1 Introduction 38 xi

15 3.2 System Design of Human Head Detection S\slem 3S 3.3 Image Acquisition and Uploading Image Pixels Declaration Imagery Buffer Declaration Arrangement of Pixel Value in the Imagery Buffer Pre-processing Module Conversion of a RGB Image into Grayscale Image Noise Removal Feature Extraction Module Edge Detection and Boundary Features Conversion of a gray scale image into BW image Detection Module Head Crux Approximation Foreground Estimation Recognition Module Bound Human Head Position System Development of Human Head Detection System Development Tools Supporting Tools Project's Graphical User Interface 71 4 RESULTS AND DISCUSSION 4.1 Introduction Pre Processing Analysis Result for Conversion of a RGB Image into Grayscale Image Result for Noise Removal Discussion for Pre Processing Analysis Feature Extraction Analysis Result for Edge Detection and Boundary Features Result for Conversion of a Gray Scale Image into BW Image Discussion for Feature Extraction Analysis Detection and Recognition Analysis Result for Head Crux Approximation Result for Foreground Estimation Result for Bound Human Head Position Discussion for Detection and Recognition Analysis System Performance Analysis The Graphical User Interface (GUI) Test of Head Detection System Result for Different Position of Head in an Image Result for Different Number of Heads in an Image Result for Different Number of Heads with Different Environment in an Image Evaluation of Result for Different Position of Head in an Image Evaluation of Result for Different Number of Heads in an Image 105 Xll

16 4.7.6 Evaluation of Result for Different Number of 1 leads with Different Environment in an Image Example of true detected result and misdetection Discussion for overall system analysis CONCLUSION AND FUTURE WORK 5.1 Conclusion Future work 115 REFERENCES 116 APPENDICES 121 BIODATA OF THE AUTHOR 123 xiii

17 LIST OF TABLES Table Page 1.1 Computer vision system application in real time environment The example above shows type of image format Comparison of Human Head Detection Result Test Result Test xiv

18 LIST OF FIGURES Figure Page 2.1 Orientation of the head is described in terms of yaw, pitch, and 13 roll as shown. 2.2 Graph for a sequence containing a yaw movement and horizontal 14 translation, with all other parameters remaining basically unchanged except for a slight roll. The top row shows ground truth. 2.3 Examples of the nested ring structure of human head regions Example Heads in different position (degree) Basic steps for filtering an image Examples of difference filter output (a) RGB image before 20 filtering (b) Median Filter (c) Wiever Filter (d) High pass filter 2.7 Hierarchy of shape based representation and description 21 technique 2.8 Examples of difference feature extraction output (a) RGB image 24 before feature extraction, (b) Robert Cross edge detector (c) Sobel edge detector (d) Prewitt edge detector. 2.9 Estimation graph for yaw and roll position Region (shaded) as it is transformed from (a) continuous to (b) 27 discrete form and then considered as a (c) contour or (d) run lengths illustrated in alternating colors. 3.1 Block diagram of system Overall flowchart of human head detection system The directions throughout to be from the point of view of the subject rather than the observer Upload input imagery flowchart The Digital Image (I) and the window kernel (h) respectively Pre processing module flowchart 47 xv

19 3.7 Median filter technique Feature extraction module flowchart Edge detection operation Detection module flowchart Chain code window kernels and its direction respectively, PI, P8 are name of pixels in an image The example results of using chain code algorithms Recognition module flowchart Microsoft Visual C++6.0 IDE Adobe Photoshop 6.0 interface and its window kernel Matlab IDE interface The Project Program Graphic User Interfaces (GUI) Image Conversion Result (a) Original image (b) Gray scale 74 image 4.2 Filtering image noise (a) Gray scale image (b) Image after implemented Median filter Boundary edge image (a) Median filter image (b) Image after implemented Prewitt edge detector Sample detection result for image sequence, (a) Input Image, (b) Image after edge detection using Prewitt Edge (c-f) Foreground detected for threshold value 25, 50, 75, 100 and 125 respectively Sample of detected curve head using T^ =125 for several of outdoor image sequences. 4.7 Comparison Binary and Chain Code output (a) Binary image, T=125 (b) After Chain Code algorithm Image the estimation location mark as red rectangular The output of edges comparison between regions interested pointed after chain code processing and estimation marking The output of estimation head curve using red rectangular. 88 xvi

20 4.11 The output of estimation human head using green rectangular Graphical User Interface of System Graphical User Interface of System after detection process Examples for Rotation are taken in the "untextured" room Examples for Scaling are taken in the "untextured" room Examples for Tilting are taken in the "untextured" room Examples for Rotation are taken in the "textured" room Examples for Scaling are taken in the "textured" room Examples of difference number human head. Head detection in RGB image (a,c,e,g.i) and Head detection in binary image (b,d,f,h,j) Examples of different environment in human head detection The analysis graft of Table The graph of Test The graph for analysis result figure The analysis graph of table (a, c, e) Sample input image, (b, d, f) Sample of true detected object (a, c, e, g) Sample input image, (b, d, f, h) Sample of misclassification of detected object Comparison performance using Chain Code and without Chain Code Comparison with those of other reviewed related works. 111 xvii

21 LIST OF ABBREVIATIONS 2D 3D API CCA DIB DIP GUI HPF IDE JPEG LPF MATLAB MF MSDN OS PED RGB ROI TC TIFF MCMC Window ME 2 Dimensional 3 Dimensional Application Programming Interfaces Chain Code Algorithm Device Independent Bitmap Digital Image Processing Graphical User Interface High Pass Filter Integrated Development Environment Joint Picture Expert Group Low Pass Filter Matrix Laboratory Median Filter Microsoft Developer Network Operating System Prewitt Edge Detection Red, Green, Blue Region of Interest Threshold Code Tag Image File Format Markov chain Monte Carlo framework Window Millennium XVlll

22 CHAPTER 1 INTRODUCTION 1.1 Introduction Nowadays computer capability has increased and computer vision itself has become a nucleus of many new eras. Subset from this scenario many applications based on computer vision have been developed such as head tracking, face detection and head orientation. Human visions are different from computer vision. But, human head detection is the first step of an autonomous human recognition system. It also has potential application in human-computer interfaces and surveillance systems. Computer vision has been applied to handle human job. So reliability and efficiency of any security system has increased. These areas of computer vision are considered to be the key requirements in many applications, such as security access control system and criminal identification system. To face the challenges in these discipline, numerous methods have been proposed, each has its own unique features and limitations. For computer vision, computer works on the array of 2 dimensional data known as an image. Performing analysis for these data can be a tedious task.

23 The output of recognition and detection algorithms can be used to provide the human operator with high level data. Besides, in order to yield accurate decision within a fast time and a proficient routine to search for stored image or video data. Advancement in the development of these algorithms would lead to breakthroughs on applications that use computer vision system. Table 1.1 shows some applications of computer vision. Table 1.1: Applications of computer vision system in real time environment. Real Time Application Tasks/Contributions Public and private security i. Monitoring department store, private properties and parking lots. ii. Robbery and vandalism. Commercial security i. Monitoring banks for crime prevention ii. ii. Access control system, Criminal identification Smart data mining i. Counting people entering and leaving the scene. 1.2 Problems Statement There are two main problems in head detection and estimation with static images that needed to be understood, analyzed and solved: 2

24 1. High Dimension State Space. Human upper body pose and pose estimation invoke searching in a high dimensional space with complex distribution. With static images, there is no preceding pose for initializing the search, unlike a video tracking problem. This calls for an efficient mechanism for exploring the solution space. In particular, the search is preferably data-driven, so that good solution candidates can be found easily [1], 2. Pose Ambiguity. From a single view, the inherent non-observability of some of the degrees of freedom in the body model leads to forwards/backwards flipping ambiguities [1] of the depth positions of body joints. Ambiguity is also caused by noisy and false observations. This problem can be partly alleviated by using multiple image cues to achieve robustness. Looking on previous researches, some of them just focus in template matching technique. In general this technique is very sensitive and cannot detect the location of the head if the person is not facing the camera and far from camera [2], The problem of human body parts recognition and detecting their pose in the 3D (three dimensional) space has been around for quite a while. A variety of methods has been developed to approach this complex problem. Determining human head pose is just one of many aspects of the mentioned problem [3], [4], [5], 3