Building an intelligent video surveillance solution requires more than video analytics, but a whole network of supporting hardware. Herve Fages, Managing Director-Asia Pacific for Pelco, discussed how to implement a smart solution at the Global Digital Surveillance Form, part of SecuTech Expo 2007, in Taipei, Taiwan.
For the last several years, the CCTV industry has been looking for a solution that offers better operator efficiency, improved response time while reducing storage capacity and achieves tasks humans cannot do. Intelligent video software, or IVS, is part of the current revolution that our industry is going through, moving from the tape recording to stand-alone digital recording and network recording.
As an additional tool offered to end users, intelligent video appeared as an obvious solution. However, it is key to understand that even though IVS performs tasks that human cannot, such as monitoring large number of scenes at once, it is neither a replacement for human operators nor a solution to all threats. It is simply an additional tool to make quick and efficient decisions.
For long time, earlier suppliers of intelligent video software oversold their applications, resulting in major end user dissatisfaction. The market is now more mature and customers need to clearly understand all the aspects of intelligent video software solutions before making their decisions.
Intelligent Video Components
Despite what some people try to convey to the market, intelligent video is not a software solution only! In order for intelligent video to operate properly, the end user must take into consideration all five components of the solution: The video capture device, or the eyes of the system. This is critical, since what cannot be seen cannot be analyzed. The network must be able to support high video quality footage in real time, which is especially important for centralized technologies.
Network latency, local storage or remote storage need to clearly be defined depending the application. The analysis device, or encoding technology, can either be on the edge (decentralized) or centralized to the entire system. The database should be centralized and redundant, storing all information. Finally, the GUI (graphic user interface) should be intuitive and user configurable, but easy , as this ultimately helps the end user customize the system to the application.
Buying an intelligent video solution by considering only the software or one of the components will result in limited performance. It is crucial to define how those five components integrate and interact together, and how easy it is to configure them from a central common interface.
Analytics devices can be either positioned at the edge (decentralized) or centralized, each of those options presenting their own pros and cons. While the centralized solution uses the headend or multiple PCs to analyze video, it requires all video to be run to a central point. It is generally less scalable and the analytics are done on video already encoded/decoded. Centralized analytics are much more cost-effective for small-scale applications with the ease of retrofitting an existing installation and a single point of failure.
However, it is not scalable. In opposition, a distributed analytic engine uses edge/decentralized equipment for video processing. With this setup, the video analytics are done before compression for high video quality, which in turn leads to better performance of analytics. This setup uses less bandwidth; therefore there are less network requirements. With the distributed engine, there is no single point of failure, which makes it more difficult to ensure redundancy.
There is an increased flexibility, so the engine can run different analytics depending on user needs, with a better systemization and ability to make linked alerts. It is obvious that with increasing performances of the front-end devices (encoder, camera, domes) and the miniaturization of electronic components, the decentralized solutions will be the long-term solution.
Analytic Solutions Available Today
As of today, intelligent video software, such as motion detection, object removal and abandoned objects, object counting, zone-and-line-crossing and license plate recognition, have shown excellent performance. Indeed, motion detection, as one of the early IVS analytics available on DVRs, is now able to run in severe conditions like rain, snow or heavy wind. Abandoned objects and object removal, named by some companies as "bomb detection," are more accurate by detecting real threats while ignoring normal objects in the scene.
Object counting, and zone and line crossing, are more accurate and have been enhanced with directional detections and tailgating features. In opposition, people counting in crowds, such as airports or train stations with open spaceas well as vehicle counting in a crowded scenegenerate many false alarms and cannot be used in real-life applications.
Despite some advertising, facial recognition in an uncontrolled environment, pickpocket/ stealing detection, automatic threat detection and people loitering in crowded scenes are generally instances where IVS does not work. Those IVS analytics required extremely well-defined scenes and it is key to clearly define the application, as well as the level of false alarms tolerated before any implementation.
Choosing the Right Solution
When choosing an intelligent video software system, it is crucial to clearly define your needs. What you aim to detect, what the scene conditions are and what level of false alarms are tolerated are all issues that need to be addressed by the system. In the design and decisionmaking process, it is important that your partner address all those aspects of the solution: software, hardware, installation and technical support. Indeed, looking only at the software algorithms without taking into consideration how it will integrate with the GUI, how the video will be recorded or transmitted over the network is like considering to buy a car after looking only at the onboard computer.
By taking the time to clearly analyze your needs and evaluate the total solution, you will avoid mismatches between your expectations and the solution that will be implemented. It is also key, before making any decisions, to determine how many manufacturers are involved in the installation. Indeed the solution may be sold by one system integrator who will have the responsibility to ensure the integration. However, the more players that are involved, the more challenging the integration will be, generating limited performance or features.
Finally software developmentas well as higher video quality with the arrival of HD encoding technologywill allow the development of new algorithms like object classification, camera tracking/hand over, anti-shaking or erratic behavior. Furthermore, with rapidly increasing new technologies such as new microprocessing, megapixel cameras, et cetera, existing IVS will improve performance, making some applications like facial recognition and loitering in crowded scenes more usable.
IVS is definitively ready now to incorporate future technology trends, bringing additional tools to the end users. Future applications will emerge in retail with customer behavior and traffic flow, as well as applications in the vertical markets such as health care, education and transportation. After being a long time "gadget," intelligent video software is now a reality of all security applications. IVS is a solution, not a single component. If defined and installed properly, IVS can be a strategic tool to improve performance, reaction time to critical events and increase return on investments.