Today, it seems everyone is watching us. In the modern urban setting, there are more closed-circuit television (CCTV) cameras recording life than ever thought possible one generation ago. The volume of information captured is creating a plethora of issues for cybersecurity professionals. These issues focus on capturing, storing and conducting analysis on CCTV video in real time.
Capturing video can be divided into 2 major categories. Camera systems are either analog or digital. Analog cameras pass the CCTV server raw video. Digital camera systems encode raw video into a format that is more conducive to transmitting across a traditional ethernet/Internet Protocol (IP) computer network. This discussion is limited to traditional CCTV camera systems that capture visible light. Common to both types of cameras is focal length and resolution. This determines how far and how accurate the camera can see objects and people in a frame.
Storing of video can become very expensive. If it is stored locally on a CCTV system, a storage strategy is usually designed to ensure continuity of operations. These strategies work on a grandfather-father-son approach to a backup site or media. This strategy is complimented with a circular approach for retiring the oldest. In other cases, the organization decides it does not want to keep video for more than a given time frame and deletes it when that time frame is met. Both approaches are valid and bound to the problem of limited video storage.
Analytics is an emerging capability that people want to see when working with CCTV. The number of algorithms is quite large, so I will just list a sampling of the major analytics being explored: face recognition, gait recognition, object recognition, patterns of life, motion and many more. Most vendors provide some level of analytics as part of their CCTV offering. However, often those analytics are limited to processing video captured by their product and only their product. For the analysis to be of any value it needs to be processed from all cameras in real time. Today, most CCTV video processing is in response to a past event and determining which cameras recorded what and when. Following this determination, the video for that event is located and processed. This is a very time-consuming and time-late business process.
Recently, I was engaged to work at a CCTV challenge involving more than 20 CCTV vendors and 10 terabytes (TB) of data. This event reinforced issues we have been tracking for as long as I can remember. Some of these issues are a result of the CCTV vendors and some of them are technical. The following are some insights that might be useful for practitioners.
The biggest issue is that CCTV vendors tend to use proprietary protocols on top of IP and use proprietary storage (under the guise of security) to protect the data and lock you into their brand. It should be noted that yes, they have an export utility, but that does not help the real-time processing paradigm. This is a hard problem to solve when anything you might write could be considered a copyright infringement with the vendor. You will not see any open-source codecs (video coders and decoders) for proprietary formats. You could replay the video and record the raw video and then process the raw video. This would take a long time and is not feasible for large CCTV architectures. The easiest way around this is to procure a software codec for their product and require the ability to tap into the video stream in real time. This would allow transfer of your data into analytic engines.
Real-time video analytics is a growing industry that promises a plethora of solutions. However, as mentioned, their problem is consuming video. To overcome this issue, organizations can just buy the analytics that are provided by their CCTV vendor. This is a valid approach for a small to medium-sized organization. However, for large organizations this becomes an interesting problem. Often, depreciation and technology refresh get complimented with competing requirements that create a wide variety of CCTV vendors. These requirements can result in several CCTV vendor who are using analog cameras, IP cameras, various protocol and storage and different levels of camera resolution and focal distances.
From a technical perspective there are several issues that you should be aware of after ingesting video and it being presented to the video analytics:
- CCTV video tends to suffer from inexpensive cameras with short focal lengths and low resolution. This means people and objects on the move tend to appear pixelated or blurred. This requires frames to be enhanced prior to presenting the frame for analytics.
- A great majority of vendors do not process video. They process frames that the analytics processes as a picture. In some instances, the analytics only process a given number of frames from the video. The analytic will sample and process every 60th or every 160th frame. The sample rate is often configurable, but the less frames you skip, the more processing performed.
- For those analytics that are processing images, vs. video, every pixel is being processed in relation to every near pixel. This means some algorithms do millions of operations to process one frame. As a result, one might want to understand the processing requirement for the analytic or the shortcuts that were taken so that the processing environment has the ability to process video. This processing requirement can be reduced by processing encoded videos based on the encoded control information.
- Artificial intelligence (AI), specifically deep learning-based analytics, is a probabilistic study. These algorithms are not giving the user absolute decisions. Instead, they are saying this the best fit to whatever criteria the algorithm is trained against. As a result, there is an open-ended issue with determining what AI analytics miss in large volumes of CCTV video.
These are just some of the high-level issues associated with video and CCTV. Ingesting videos for an integrated operational picture (pun intended) across a multivendor CCTV architecture can become difficult and frustrating. Plan your CCTV procurement from the perspective that you may need to move out of the vendor’s architecture and process video in real time on another platform. In the end, the solutions for CCTV are just coming of age. As the demand signal rises for more integrated CCTV industry standards and common formats, most of these issues will cease to exist.
Bruce R. Wilkins, CISA, CRISC, CISM, CGEIT, CISSP, is the chief executive officer of TWM Associates Inc. In this capacity, Wilkins provides his customers with secure engineering solutions for innovative technology and cost-reducing approaches to existing security programs.