Categorising the Abnormal Behaviour from an Indoor Overhead Camera A PROJECT REPORT Submitted in partial fulfilment of the requirement for the award of the degree of BACHELOR OF TECHNOLOGY IN ELECTRONICS AND INSTRUMENTATION By GURKIRT SINGH (06BEI080) Under the Guidance of Dr. R. B. Fisher Dr. P. Arulmozhivaraman University of Edinburgh VIT University SCHOOL OF ELECTRICAL ENGINEERING VIT University VELLORE – 632014, Tamil Nadu, India June 2010 ACKNOWLEDGEMENT I am grateful to my Division leader and the Director of SELECT, for giving me the permission to carry out this project. I would like to sincerely thank Dr. Bob Fisher for giving me the opportunity to carry out my project work in the School of Informatics at University of Edinburgh. I express my heartfelt and deepest gratitude to Dr. Bob Fisher for his guidance through the completion of the project and for providing us with the necessary inputs. I would also like to thank Steven and Michal for their help and support. I would like to thank my guide, Dr. P. Arulmozhivarman for his guidance and co-operation. I extend my gratitude to all the teachers who have ever taught me. Without the knowledge that they imparted over the 7 semesters, completion of this project would not have been possible. Finally I want to thank my parents for their help and supporting morally and financially to carry out my work in Edinburgh. Gurkirt Singh (06BEI080) ii To my parents and family My father Mr. Rakha Singh and mother Mrs. Gobind Kaur and my Dear nephews and niece. iii Declaration I declare that this thesis was composed by myself, that the work contained herein is my own except where explicitly stated otherwise in the text, and that this work has not been submitted for any other degree or professional qualification except as specified. iv Abstract This dissertation describes a B.Tech project for which the purpose was to develop a system that could be used for automated surveillance. The main novelty is the use of a vertical camera. The project investigates whether such a system can effectively detect the moving objects, track their trajectories, and use these to recognise anomalous events. A vertical camera is used to capture continuous video for detection, which involves low level of image processing to detect the objects and stores their properties and positions. The tracking program is used to track the person and form a trajectory from the detector output file. The tracking program produce the file of tracked persons trajectories, all trajectories have different lengths depending upon how much time the person stays in view of camera. Normally a person stays in view of the camera for 10 to 15 seconds. To describe each trajectory with an equal number of attributes Spline fitting algorithm is used, which gives the six control points for each trajectory. The detection of anomalous behaviour is described by different number of parameters like error in spline fit, vector distance from closest mean vector and multivariate Gaussian probability. v Table of Contents i. List of Figure ...................................................................................... vii ii. List of Table ......................................................................................... ix Chapter 1 Introduction ...................................................................................... 1 1.1 Motivation ................................................................................................... 1 1.2 System Objective ........................................................................................ 2 1.3 The Forum ................................................................................................... 2 Chapter 2 Background ...................................................................................... 4 Chapter 3 System Overview .............................................................................. 5 3.1 Detection ..................................................................................................... 5 3.2 Tracking ...................................................................................................... 8 3.3 Representing Trajectories as Spline ........................................................... 8 3.4 Abnormality Detection ............................................................................... 8 Chapter 4 Tracking ............................................................................................ 9 4.1 specification ................................................................................................ 9 4.1.1 Merging and Splitting ......................................................................... 12 4.1.2 Disappearing ...................................................................................... 13 4.1.3 People as Separate Blobs .................................................................... 13 4.2 Design ....................................................................................................... 13 4.2.1 Corresponding Person and merging ................................................... 13 4.2.2 Splitting ............................................................................................. 15 4.2.3 Splitting ............................................................................................. 15 4.2.4 People as Separate Blobs ................................................................... 15 4.2.5 Core Algorithm ..................................................................................... 16 vi 4.2.6 Calculations ...................................................................................... 22 4.2.6.1 Velocity ................................................................................... 22 4.2.6.2 Predict Position ....................................................................... 22 4.2.6.3 Radius ..................................................................................... 23 4.2.6.4 Probabilities ............................................................................ 24 4.2.6.4.1 Prediction Error Probability .......................................... 24 4.2.6.4.2 Histograms Probability ................................................. 25 4.2.6.4.3 Angles Probability ........................................................ 25 4.3 Removing Bad Trajectories ..................................................................... 26 4.4 Implementation ......................................................................................... 29 4.5 Output File Description ............................................................................ 30 4.6 Evaluation ................................................................................................. 30 Chapter 5 Representing Trajectories as Spline ............................................ 35 5.1 Specification ............................................................................................. 35 5.2 Design ....................................................................................................... 35 5.2.1 Choosing Control Points ..................................................................... 35 5.3 Implementation ......................................................................................... 37 5.4 Output File Description ............................................................................ 37 Chapter 6 Abnormality Detection .................................................................. 39 6.1 Overview ................................................................................................... 39 6.2 Building the Model ................................................................................... 40 6.2.1 Calculations ........................................................................................ 41 5.3 Choosing Parameters ................................................................................ 44 5.4 Implementation ......................................................................................... 51 5.4 Results ....................................................................................................... 51 vii Chapter 7 Conclusions and Future Work ..................................................... 53 References ......................................................................................................... 54 List of figures S.No. Index No. Name Page No. 1 1.1 The view from the camera. 2 2 1.2 Entry and exit points 3 th 3 3.1 Detected points on 4 January 2010 7 4 4.1 An example of two persons merge 10 5 4.2 An example of two person split 10 6 4.3 An example of a person disappearing and 11 another person appearing at the same time 7 4.4 An example of one person being detected as 11 several blobs 8 4.5 A simple scenario of people walking together 12 9 4.6 Bhattacharyya distances between histograms 14 of the same and different people. 10 4.7 The distribution of prediction errors 25 11 4.8 The marginal area where trajectories have to 28 start and end. 12 4.9 Block Diagram of Tracker program 29 13 4.10 Single person trajectory 31 14 4.11 Example of merging and splitting condition 31 15 4.12 Example of disappearing 32 16 4.13 Tracked object for 4th January 2010. 33 17 4.14 Image plot of all the detection points after 34 tracking using all trajectories (65529). 18 5.1 Variation of Median fitting error with number 36 of control points viii th 19 5.2 Spline curves for 4 January 2010 38 20 6.1 An example training cluster 40 21 6.2 Variation of number with the variation of 41 spline fit 22 6.3 Variation of vector distance to closest mean 42 vector. 23 6.4 Variation of numbers on each log(probability 43 24 6.5 Variation of false positive and false negative 45 at different values of spline fit error threshold 35 6.6 Variation of false positive and false negative at 46 different values of spline fit error threshold 36 6.7 variation of false positive and false negative at 46 different values of log(probability) threshold 37 6.8 Examples of false negative trajectories 47-48 38 6.9 Examples of false positive trajectory 48-49 39 3.10 Clusters those are modelled (107) 51 40 3.11 Shows four different clusters for different 52 paths List of Tables S.No. Index No. Name Page No. 1 4.1 The failure rate of correct identity 32 2 6.1 Shows equal error rate for each parameter 47 3 6.2 Thresholds values at best equal error rate for 47 training. 4 6.3 Shows number of trajectories in all 169 49 clusters present from total 65529 trajectories ix