The Metropolitan Police in London have found that the opportunity to use computer vision technology in the analysis of real-world street-scene video is severely limited because of the practical constraints in the variety and poor quality of videos available to them. Consequently, in a large criminal investigation, police forces employ numerous officers and volunteers to watch many hours of camera footage to locate, identify and trace the movements of suspects, victims, witnesses, luggage and other inanimate objects. Their goal is to piece together a story of events leading up to an incident, and to determine what happened afterwards. In this paper, we present the technical challenges facing researchers in developing computer vision technique to process from the wild street-scene videos.

Finding corresponding key points in images from security camera videos is challenging. Images are generally low quality and acquired in uncontrolled conditions with visual distortions caused by weather, crowded scenes, emergency lighting or the high angle of the camera mounting. We describe a methodology to match features between images that performs especially well with real-world images. We introduce a novel \emph{blur sensitive feature detection} method, a combinatorial feature descriptor and a distance calculation that efficiently unites texture and colour attributes to discriminate feature correspondence in low quality images. Our methods are tested by performing key point matching on real-world security images such as outdoor CCTV videos, and we demonstrate an improvement in the ability to match features between images compared with the standard feature descriptors extracted from the same set of feature points. We use key point features from Harris Corners, SIFT, SURF, BRISK and FAST as well as MSER and MSCR region detectors to provide a comprehensive analysis of our generic method. We demonstrate feature matching using a 138-dimensional descriptor that improves the matching performance of a state-of-the-art 384-dimension colour descriptor with just 40% of the storage requirements.

Our research is concerned with the practical application of computer vision in the forensic analysis of security images and videos. Contemporary literature make use of high-definition images and Hollywood feature films in their datasets, and there is little or no assessment of algorithms' performance using poor quality images with variable frame rates and uncontrolled lighting conditions such as security video.

Work so far has produced a methodology for matching features across low quality images that yields an improved results over existing feature matching techniques. Future work will involve innovation in search and retrieval online machine learning to train models from unlabelled data, and segmentation of one-shot videos to aid computer and human analysis of long-running video sequences. We are motivated to produce an integrated system for police investigators to use a query-by-example search and retrieval system with relevance feedback and machine learning to incrementally discover evidence in criminal investigations.

@inproceedings{Henderson2014,
address = {Nottingham},
author = {Henderson, Craig and Izquierdo, Ebroul},
booktitle = {6th Doctoral Consortium at BMVC 2014},
title = {{Large-scale forensic analysis of security images and videos}},
year = {2014}
doi = {10.13140/2.1.2859.0887},
url = {http://dx.doi.org/10.13140/2.1.2859.0887}
}