Author: Dr Mark Sugrue, founder and chief technology officer, Kinesense Ltd  'Big Data’ is creating big changes across society. The term is loosely defined, but at its heart is the idea of mining voluminous ‘unstructured’ data and to produce useful answers. Supermarkets use big-data techniques on their huge customer transaction database to predict what people want to buy or to send them coupons and offers. US retailer Target uses customers’ purchase history to predict which of its customers are pregnant and when they are due – and sends them promotional coupons for baby products (1). The GAA has started storing anonymised performance data on all players and using big-data techniques to help teams improve training (2). The volume of these kinds of datasets is eclipsed by closed-circuit television (CCTV). Police forces in every country are some of the biggest data users, with CCTV video the largest body of unstructured data. The quantities of CCTV are staggering. In the UK alone, there are somewhere between two million and five million CCTV cameras in use, generating 25 billion hours of recorded video per year. In data terms, this amounts very roughly to 25 exabytes (3) – or about 25 million terabyte hard disks. For comparison, this is about two million times more data than is stored in all the books held in the US Library of Congress. Of course, most CCTV data is discarded and never watched or processed by anyone. Typically, CCTV is recorded and stored for one or two weeks before being overwritten. When recorded CCTV is watched, it is often by the police after a crime has been reported. Again, in the UK (which has the best kept statistics on this subject), some 64% of their two million criminal cases per year involve CCTV evidence. At Kinesense, we work closely with police forces in Ireland, the UK and a dozen other countries, to improve systems for dealing with CCTV workloads. Surveys we have carried out (4) indicate that the average amount of video per case is 30 minutes. However, there is a wide variation. For a smaller number of major investigations, thousands of hours of CCTV can be collected. There are a few cases each year, chiefly missing-person or abduction cases, where 100,000 hours of CCTV are collected from thousands of cameras across a wide area. Police must review this video to find clips of the crime itself, or the suspect travelling to or from the scene. They will store all this data for years as the case progresses though court and appeals process. Overall, over 15% of all UK police time is taken up with collecting and using CCTV. The figure for Ireland is similar.

Police CCTV and Big Data

There is a lot of scope to make this more efficient using the latest technology, saving money for the tax payer and, importantly, putting those limited police resources to better use in crime prevention. Based in Dublin’s Merrion Square, Kinesense was founded to tackle the CCTV Big Data challenge. The solution we came up with was a technology for searching the content of video, indexing it like a search engine and making the whole thing searchable in an intuitive way. The key is to index all the activity in the video as it happens – people walking, cars driving, colours, directions, etc – and let the investigator define his query after the fact. Usually, an investigator does not know what he is looking for from the beginning of the investigation, and may need to look back at the same video when new information is needed. You do not want to have to re-index the video when the search query changes. One of our police customers told us of how they used our software to search hundreds of hours of CCTV to follow the movements of a drug dealer who was visiting a safe house. In questioning later, he claimed he had not met a particular co-accused at the house. One of the officers quietly nipped out of the interview room, searched the video again and found several clips of the co-accused visiting the house and meeting the suspect. Within minutes, the officer was able to return to the interview with new evidence in hand. Before this technology was available, rewatching the video would have taken days at a minimum and often the manpower would not be available to do it at all. The indexing algorithm is the heart of the solution and it needs to achieve two opposing aims. Firstly, it must be extremely sensitive and accurate – detecting the smallest important motion in the video and recording it correctly in the search index and missing nothing important even when the video quality is poor. Secondly, it must be insensitive to noise and spurious motion from, for example, tree branches moving in the wind or flickering lights. We manage this by using an element of machine learning that adapts to activity in the scene. For example, if part of the scene is ‘noisy’ due to a branch blowing in the wind, the algorithms can map this motion pattern and screen it out – yet a person walking past the tree can still be detected because their motion signature is different from the tree. Similarly, rain or lighting effects are automatically ignored (5). Our police customers report that using our technology saves them 95% of the time it used to take to search and review video. They find it more accurate than manually watching video, too – because humans are just not good at staying attentive during very long, and quite boring, CCTV videos. It is a task that, by their nature, computers can do better. Our technology turns the ‘big’ unstructured data of CCTV into useful answers or ‘intelligence’. Of course, generating the ‘intelligence’ is only the first step, and only useful if the police can share it easily with colleagues, deliver it to the prosecution and defence teams, and send it to the court system. Kinesense has teamed up with UK multinational Northgate Public Services to achieve this. Northgate have created the GEM3 framework for sharing data between police and the public prosecutor IT systems. This creates a comprehensive workflow – unstructured raw data, to intelligence, to evidence. The efficiency savings to the police and ultimately the tax payer are considerable.

CCTV technology and private companies

Many private companies face the same CCTV challenges as police forces. We work with telecoms to cut copper thefts, a serious issue faced by all large infrastructure companies. The challenge is not cost or budget restrictions – such companies spend considerable sums on CCTV cameras. The challenge is turning that vast quantity of raw video data into useful information as quickly as possible. Likewise, large distribution companies who operate thousands of cameras monitoring their storage and sorting areas also use our solution. The future for Kinesense is bringing our technology to a wider global market, and that means integrating with existing systems. IBM is a big player in cloud services and Big Data and we recently tied in with their i2 platform. i2 is used by police, insurance companies and others to bring together disparate types of data and search it for unexpected connections. For example, for police monitoring a drug gang, CCTV intelligence from Kinesense can be seamlessly searched along with phone records and credit-card histories to uncover hidden relationships and suggest new lines of enquiry. The number of CCTV cameras is increasing year on year. They are being joined by the personal wearable cameras like Google Glass (GoPro is already common) and police body-worn cameras. The result will be an explosion in the amount of data that needs to be processed when an investigation is needed. For police or any organisation that faces being swamped by too much unstructured data, Kinesense keeps their head above water. References: (1) 'How Target Figured Out A Teen Girl Was Pregnant Before Her Father Did'. Forbes, Kashmir Hill, 16/2/2012, http://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/ (2) 'Big Data’s big impact on the GAA.' Sunday Business Post, July 2014 (3) Exabyte – Wikipedia http://en.wikipedia.org/wiki/Exabyte (4) Kinesense Survey of Volume Crime CCT V, 2014 http://www.kinesense-vca.com/2014/03/volume-crime-cctv-survey-results/ (5) Example of tracking someone walking behind branches: http://youtu.be/36Y9bLbIc6c Dr Mark Sugrue is founder and chief technology officer of the video analytics company Kinesense Ltd and is responsible for new product development. He holds a BSc in applied physics from the University of Limerick and a PhD in video analytics from the Royal Holloway, University of London. Kinesense has worked with law enforcement and security agencies in Europe and North America for many years and is focused on making CCTV more efficient and easier to use for the police. Kinesense products and services include the video search and reporting products ‘Kinesense Law Enforcement’ and ‘CovertSuite’, the free Vid-ID.com tool and player manager.