CCTV PhD

Information-Physics One

Special Reception for Younger Physicists

Company making car safety systems. Jobs

SCAMP: a general-purpose focal-plane massively parallel processor array Chip which analog detects motion.

full_192.pdf (application/pdf Object)&bmvc2001.pdf (application/pdf Object)Tim Ellis and D Makris on detecting and analysing the path of pedestrians.

ScienceDirect - Artificial Intelligence : Visual surveillance in a dynamic and uncertain world*1 Sicence Direct - login

Psarrou-Gong-Walter-IVC2002.pdf (application/pdf Object) Recognition of Human gestures and behaviour based on motion trajectories.

CCTVpaper.pdf (application/pdf Object) What happens next? The predictability of natural behaviour
viewed through CCTV camerasÀ
Perception, 2004, volume 33 ages 87 ^ 101
Tom Troscianko, Alison Holmes, Jennifer Stillmanô, Majid Mirmehdi, Daniel Wright½,
Anna Wilson½
Department of Experimental Psychology, University of Bristol, 8 Woodland Road, Bristol BS8 1TN, UK;
ô School of Psychology, Massey University, Albany, North Shore MSC, Auckland, New Zealand;
½ Department of Psychology, University of Sussex, Falmer, Brighton BN1 9QH, UK;
e-mail: tom.troscianko@bris.ac.uk
Received 1 July 2002, in revised form 15 September 2003
Abstract. Can potentially antisocial or criminal behaviour be predicted? Our study aimed to ascertain (a) whether observers can successfully predict the onset of such behaviour when viewing real recordings from CCTV; (b) where, in the sequence of events, it is possible to make this prediction; and (c) whether there may be a difference between naive and professional observers. We used 100 sample scenes from UK urban locations. Of these, 18 led to criminal behaviour (fights or vandalism). A further 18 scenes were matched as closely as possible to the crime examples, but did not lead to any crime, and 64 were neutral scenes chosen from a wide variety of noncriminal situations. A signal-detection paradigm was used in conjunction with a 6-point rating scale. Data from fifty naive and fifty professional observers suggest that (a) observers can distinguish crime sequences from neutral sequences and from matches; (b) there are key types of behaviour (particularly gestures and body position) that allow predictions to be made; (c) the performance of na|« ve observers is comparable to that of experts. However, because the experts were predominantly male, the absence of an effect of experience may have been due to gender differences, which were investigated in a subsidiary experiment. The results of experiment 2 leave open the possibility that females perform better than males at such tasks.
DOI:10.1068/p3402

Human Vision Models for Perceptually Optimized Image Processing - A Review

Marcus J. Nadenau, Stefan Winkler, David Alleysson, Murat Kunt

Human Visual System, Color Perception, Image Compression



Person Identification Using Multiple Cues

Roberto Brunelli, Daniele Falavigna

IEEE Transactions on Pattern Analysis and Machine Intell. V17 No. 10 oct 95

Face Recognition. Not very useful for my project.



On Illumination Invariance In Color Object Recognition

M Drew, J. Wei, Ze-Nian Li

Linear Color Algorithm



Mobile Robot Localization Under Varying Illumination

M Jogan, H Wildenauer H Bischof

Eigenspace EigenImage

-each learned image is point in low-dim eigenspace. Project scene on eigenspace and search for closest learned point. Uses PCA to reduce the dimentionality of an image to plot as a point. Method is robust with occulusions and illumination artifacts but global Illumination must be delt with by filters.



Robustly Estimating Changes in Image Appearance

Black, Fleet, Yacoob

appearance change, optical flow. Review of multitude problems of image change.



Illumination Invariant and Occlusion Robust Vehicle Tracking by Spatio-Temporal MRF Model

Kamijo, Sakauchi

Spatio-Temporal Markoff Random Field.





An Illumination Invariant Change Detection Algorithm

Lou Yang Hu Tan

Seems useful but difficult to understand. Color. Divides image into stationary background and moving foreground. homomorphic filtering for Ill, invar. Internet sites on this method say it is very slow.



Illumination-Invariant Image Retrieval and Video Segmentation

Drew, Wei, Li.

A lot of emph on Color.



Adaptive Background Estimation: Computing a pixel-wise learning rate from local confidence and global correlation values

PIC, Berthouze, Kurita 2004

Very interesting. Results in robust moving object segmentation.



A Novel Background Initialization Method in Visual Surveillance

Bevilacqua

Bayes Theorm. looks at each pixel as it changes through time. Picks most usual intensity and calls it background, under the assumption that moving foreground objects pass by quickly and background intensities are dominant in time. Uses this to build a background model and subtracts from image to get moving objects. Doesn't deal with background illumination changes in time but says method robust for all illumination conditions (assuming they are constant in time)



Effective Object Segmentation in a Traffic Monitoring Application

Bevilacqua

3x3 morph method for segmenting video objects.



Image Segmentation by three-level thresholding based on maximum fuzzy entropy and genetic algorithm.

Tao, Tian, Liu

probability partition. Extends work of Zhoa et al 2001, IEEE Trans. Fuzzy Systems 9 (3), 469-479



Local Entropy-based region extraction and thresholding

Yan, Sang, Zhang

Local Entropy, Transition Region, Gradient, Segmentation.



Phd: Real-Time Occupant Detection in High Dynamic Range Environments - Carsten Koch

		Journal Review

Optical Dynamic Range  - Yamada: effectivness of video camera dynamic range expansion for lane mark detection, IEEE Int. Conf on Intell. Vehicles, 1998 -> CCD cameras fail 10% of time to detect lane marks in daylight



Usuall Dynamic Range of Cameras is 60-70dB. High Dynamic Range cameras for use in extreme lighting envoron use CMOS and give 80dB



Background Segmentation. When an object moves it joins the foreground but fades back to the background shortly after stopping.



Whole object should be moved to foreground - not just it's moving components.



---------------------------------------------------



Watkinson "The MPEG HandBook"

-motion compensation - optical axis in 3d (2d space, 1 time) space. optical axis at angle to time axis if object moving. object changes little as it moves along axis. (MPEG-4)



Stauffer, Grimson, "Learning Patterns of Activity Using Real-Time Tracking"

realtime visual tracking, adaptive background estimation, gaussian, surveillance.

107. S. A. Niyogi and E. H. Adelson, Analyzing and recognizing walking figures in XYT, in Conference on Computer Vision and Pattern Recognition, 1994. One of the very few papers that use XYT processing of video.

cvpr,1997 [113] Pedestrian Detection Using Wavelet Templates, IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'97)

This paper presents a trainable object detection architecture that is applied to detecting people in static images of cluttered scenes. This problem poses several challenges. People are highly non-rigid objects with a high degree of variability in size, shape, color, and texture. Unlike previous approaches, this system learns from examples and does not rely on any a priori (handcrafted) models or on motion. The detection technique is based on the novel idea of the wavelet template that defines the shape of an object in terms of a subset of the wavelet coeficients of the image. It is invariant to changes in color and texture and can be used to robustly define a rich and complex class of objects such as people. We show how the invariant properties and computational eficiency of the wavelet template make it an effective tool for object detection.

A multiscale approach to connectivity General discussion of Multiscale connectivity
(pdf)

[25] O. Chomat and J. L. Crowley, Recognizing motion using local appearance, in International Symposium on
Intelligent Robotic Systems, University of Edinburgh, 1998.

This paper presents a technique for the analysis and recognition of body motion using local appearance. A set of spatio-temporal lters are determined by principal component analysis of the contents of spatio-temporal neighborhoods from a set of image sequences. A few principal components dene the basis for an orthogonal space for describing the appearance of motion. The axes of this space represent local spatio-temporal patterns. The projection of a local spatio-temporal neighborhood onto these axes provides a vector which describes the neighborhood. Motion can be described statistically using the multi-dimensional histogram of projections. Such statistical analysis of feature projections provides the basis for a method for recognition of motion patterns. In this paper we report on initial experimental which show promising results in the description of a walking action