Intelligent camera systems are on the way

As cameras become more common, security’s job gets harder. Spotting a suspect who tries to pull out a gun from his pocket may be an easy job for a well-trained security personnel. However, finding enough resources to control and monitor activities captured by hundreds of cameras, and making the correct decision in limited time and under pressure can be costly. In the near future, smart camera systems will take at least part of the load from human beings. However, smart cameras will not replace people, but help them to work more effectively. Machine vs. man Extensive research has been devoted to prove that machine vision systems are superior to the human eye in certain applications that require constant monitoring and comparison with a large database in real-time. However, early software systems for vision had many problems: they were hard to scale, had poor real-time performance and were hard to program and extend. On the other hand, recent advances in camera and storage systems has driven increased popularity of multi-camera systems. Recently, the market interest started to increase, thanks to the success of the machine vision integrated camera system solutions. The applications include label inspection, assembly verification, quality testing, optical character recognition, customer behavior analysis, security surveillance and identification. Computer vision is being more widely used in visual surveillance applications because demands for coverage and quality have become so stringent. Current security and surveillance systems depend on human operators’ constant attention to detect any suspicious activity. However, with the increased number of sensors deployed in large and active environments the need for security personnel increases significantly. While dumb cameras require people to monitor, verify, process and interpret, smart cameras capture high-level descriptions of the scene and analyze what they see. These devices can support a wide variety of surveillance applications including enhanced security by discovering suspicious patterns of activity, noticing unattended objects, identifying intruders, and detecting hostile operatives near critical facilities. These technological advances will create new security possibilities but also new privacy and other social and ethical issues. Two classes of solutions From the hardware point of view, the solutions can be categorized into two classes, namely PC-based vision systems and integrated smart camera systems. The major drawback of the integrated smart cameras is finding the right application and designing and optimizing the camera in terms of power, size, etc., under multiple application constraints. We believe that the basic step in integrated smart camera design is the workload characterization that is very much dependent on the application. PC-based intelligent camera systems can be easily modified for new applications and support very complex and power consuming algorithms. The experts say Joseph P. Freeman, from J.P. Freeman Co., says smart cameras on the market today only have pieces of intelligence, such as object tracking, according to his 2003 Worldwide CCTV & Digital Video Surveillance Market report. John Devlin, from IMS Research, claims that although sales of new product types, such as integrated smart cameras and smart sensors, continue to grow, they won’t replace PC-based solutions. Prices continue to drop on components, e.g. CMOS cameras, while manufacturers have added more features. Furthermore, the evolution of digital video, especially in digital video storage and retrieval systems, increases the popularity of video surveillance. The result of this development effort will fill the gap between powerful hardware and primitive software in today’s commercial surveillance systems. These topics are still open areas for many research groups in industry, government and academia. Burak Ozer is the co-founder of Verif-icon Corp. and can be reached at ozer@verificon.com. Wayne Wolf is the co-founder of Verificon Corp. and professor at the Department of Electrical Engineering, Princeton University. He can be reached at wolf@verificon.com.