|
It won`t be long before we all benefit from intelligent performance surveillance systems that detect far more than the human-eye. Improvements in both surveillance systems and back end intelligent analysis software have evolved to the point where intelligent performance surveillance systems are not just possible, but also practical. This article will explore recent advances in surveillance systems and describe how improved image quality enables intelligent performance surveillance.
By Jean-Pierre Forest
CCTV HISTORY AND IMAGE QUALITY
Prior to the early 80’s when VCRs took over the surveillance industry, most banking establishments were equipped with Friscoe Bay type 35mm film cameras that were triggered by a teller during a robbery. After the incident, the police would recover the film, have it developed, and then they would have high quality 8 X 10 photographs of their suspects. Although the image quality was excellent, the main problem with using 35mm film was that it was time consuming to develop the negatives, and print the images, which is why the industry moved to video cameras and VCRs.
Soon after, time lapse VCRs were introduced. By slowing the tape in the VCR cassette, less frames per second were recorded, but recordings could span up to 40 days (960 hrs). One of the problems with Time Lapse VCRs is that the VHS format could only record a maximum of 280 lines of resolution and could only take one video input. To record more than one camera to a VHS tape, multiplex coder/decoders and video switchers were then used to increase the amount of cameras on one video tape. This further reduced the amount and quality of images recorded by the time lapse recorder.
In the late 1990’s, the security industry started deploying Digital Video Recorders (DVRs) which took analogue camera inputs and recorded the images using lossy compression onto a hard disk. This approach to storing surveillance footage was now taking an already bad image and making it worse. DVR manufacturers extolled the virtues of their ‘digital’ systems advertising large compression ratios and small file sizes per image. Manufacturers of DVRs advertise file sizes down to 2.2Kb, however, if a normal video image has a digital matrix of 720 X 486 pixels, the uncompressed storage of this full image would normally be around 1Mb in size which is 1,000Kb. So the following question must be raised. Where is the 998Kb of data that was discarded in the ‘Lossy’ compression recordings? In spite of the poor image quality DVRs offered, the security industry quickly adopted them due to the convenient features they offered that were not possible on tape-based systems such as easy searching and scheduling.
By early 2000, image quality in surveillance systems was getting so poor, that the Scientific Working Group on Imaging Technology (SWGIT) collaborated with other scientific working groups to address imaging concerns raised by law enforcement officials and security professionals created by the proliferation of poor quality CCTV systems. In 2004 SWGIT released “Recommendations and Guidelines for Using Closed-Circuit Television Security Systems in Commercial Institutions”. These guidelines are instrumental in addressing the concerns of law enforcement agencies and security professionals relating to the use of CCTV systems. A copy of these guidelines can be obtained from section 4 of the International Association for Identification website at:
http://www.theiai.org/guidelines/swgit/index.php.
THINKING OUTSIDE THE BOX
To improve the image quality to the level required for security professionals and law enforcement agencies, the industry has had to completely rethink how it engineers surveillance systems. As our knowledge of imaging, digital transmission, storage and viewing technologies increases, we have taken great strides towards delivering the promise of Intelligent Performance Surveillance. The biggest technical improvements to date have been on the front end where almost every aspect of image gathering has improved. A few of the most important developments are detailed below.
IMAGING: CAMERA IMPROVEMENTS
Today, for the first time, it is feasible to deploy multi-megapixel cameras that use progressive scan scientific-grade CCD sensors and that deliver true 14-bit dynamic range up to 1,100nm. Extremely sensitive across the entire visible spectrum and into near infrared wavelengths, the latest surveillance cameras engineered with scientific grade sensors see beyond what the human eye can see. The human eye can produce about 8-bit dynamic range and sees only to 700nm, so performance surveillance cameras see well beyond human vision (they can actually see 64 times more dynamic range) and are far more sensitive than the best NTSC-based traditional cameras which provide about 8-bit dynamic range.
In addition, scientific-grade sensors have higher Quantum Efficiencies (QE) than their consumer-grade counterparts. On average consumer-grade sensors found in typical surveillance system have a QE of about 15% while scientific-grade sensors have a QE of up to 60%. This allows scientific sensors to capture up to 3 times more information than conventional surveillance cameras that use consumer-grade sensors. This new generation of surveillance cameras can generate images with resolutions of up to 11 megapixels. This level of detail permits the use of digital Pan/Tilt/Zoom on recorded footage to extract image details after an incident has occurred which is impossible to do with conventional PTZ systems.
|
|
|
|
Source: Avigilon
|
|
|
|
|
Performance Surveillance Image (left)/ Traditional Surveillance Image (right) Source: Avigilon |
EFFICIENCY OF A CAMERA SENSOR
Another recent improvement has to do with the ability of cameras to change their image acquisition parameters as lighting conditions change. Ambient lighting often changes and this presents a major challenge to the reproduction of high quality images. High performance cameras today can be equipped with the ability to analyze a scene in real-time and adjust the camera’s settings accordingly. For example, if exposure time is too short, not enough light energy reaches the sensor and the signal-to-noise ratio goes down which can make an image appear ‘noisy’ or ‘speckled’. If an exposure time is too long, the signal to noise ratio may be fine, but moving objects in the scene end up blurry, known as motion blur.
As light in a scene drops, the camera should make adjustments to exposure time and aperture to achieve an optimal signal-to-noise ratio while still maintaining image sharpness for moving objects and optimizes the depth of field. Performance surveillance cameras available today capture the best possible image for the given conditions by continuously adjusting exposure time, lens aperture size and signal-to-noise ratio.
REDUCING COMPRESSION ARTIFACTS
Once an image has been captured, the next step is to transmit that information to storage and viewing facilities. In the past, because there were severe bandwidth limitations, conventional surveillance recorders used ‘Lossy’ compression techniques that introduce image artifacts, and discard potentially important details. Surveillance camera manufacturers tend to emphasize the benefits of their specific lossy compression technique, instead of acknowledging the fact that these methods permanently destroy critical image data. Most lossy compression algorithms in use today were designed for the entertainment or video conferencing industry not for surveillance where attention to detail is crucial. MPEG1, MPEG2, MP4, h.263 may be fine when used in the entertainment applications for which they were designed, however, if it is necessary to identify the person who abducted a child, law enforcement and security professionals do not want images compressed with lossy compression, they want high quality images with the maximum amount of information available to identify the perpetrator. Fortunately, there is an alternative to lossy compression. Lossless compression standards preserve original image data while still reducing the required bandwidth and storage space. To ensure that images are of the highest quality and to transmit and store surveillance evidence in an efficient manner, performance-surveillance systems must use lossless compression. Although high-performance cameras that use lossless compression require more bandwidth and hard drive space, the resulting images that are stored on the recorder contain exactly the same data as when they were originally captured by the surveillance camera. One of the enabling technologies for the use of lossless compression is Gigabit transmission over Ethernet cabling. Gigabit Ethernet cameras can stream image data 10 times as fast as cameras that only support 100Mb Ethernet. The extreme transmission rates provided by Gigabit Ethernet allow high performance cameras to deliver multi-mega pixel images without being forced to resort to lossy compression techniques.
Figure 1. Both images have a format resolution of 320 X 320, however, the image on the left does not have the visual resolution or sharpness of the image on the right. (Source: Avigilon)
ENHANCED VIEWING: BETTER THAN HUMAN PERCEPTION
Using high performance cameras, Gigabit Ethernet transmission, and high capacity recorders, performance surveillance systems capture and store images exactly as the image sensor detected them. Since performance surveillance systems capture more image information than is visible with the unaided eye, image enhancement tools, such as automatic contrast and levels enhancement, are necessary to reveal their full detail.
Automatic contrast and levels enhancement re-maps the dynamic range of the image to the dynamic range of the monitor on which the image is being viewed. This ensures that the maximum contrast possible is displayed. Since the dynamic range of images from high performance cameras exceeds both that of the monitors and the human eye, regions of interest can be selected and the levels enhancement applied to that region. This allows critical details to quickly and easily be revealed. Levels enhancement is particularly useful for discerning details in dark or bright sections of an image.
RESOLUTION: HOW MUCH IS ENOUGH?
In general, to reveal detail in an image, more resolution is better. The number of pixels in an image, however, is not the sole indicator of image quality. Images from a performance surveillance system will appear much clearer and sharper than images captured with conventional surveillance systems even if the resolution of the two images is the same.
SAME RESOLUTION, DIFFERENT RESULTS
In Figure 1, both images have a format resolution of 320 X 320, however, the image on the left does not have the visual resolution or sharpness of the image on the right. To determine how much resolution is necessary for a surveillance camera, it is important to determine the role of the camera in the overall surveillance system. The Underwriters Laboratories of Canada has a standard (ULC S317-96) for the performance of CCTV systems. These standards are based in part on the following 4 roles that a surveillance camera can play in a surveillance system:
Monitor
An observer can determine the number, direction and speed of movement of people whose presence is known to him; i.e., they do not have to be searched for.
Detect
Following an alert an observer can, after a search, ascertain with a high degree of certainty whether or not a person is visible in the pictures displayed to him/her.
Recognize
Viewers can say with a high degree of certainty whether or not the individual shown is the same as someone they have seen before.
Identify
Picture quality and detail should be sufficient to enable the identity of a subject to be established beyond reasonable doubt.
|
|
|
Source: Avigilon |
SEEING MORE HOLISTICALLY
Recent advances in digital imaging technology, computers and networking hardware make it possible to usher in a new level of performance in surveillance systems. Unfortunately the business of creating quality images to power evidentiary forensics and intelligent analytics is not an easy task. Simply improving a camera sensor or the transmission method, or the viewing system is not enough. The entire system must be viewed as a whole and implemented as such.
A performance surveillance system must be designed as a whole because each precedent component must be engineered to capture, preserve and supply a complete, accurate picture to each subsequent component in order to preserve image quality down the chain. To engineer such a system requires significant effort and technical skills that many companies that focus on cameras or networking gear, simply do not have. Figure 2. shows the difference between traditional systems and performance systems in terms of image quality.
WHERE INTELLIGENT SURVEILLANCE STRUGGLES
Unfortunately, the promise of intelligent surveillance has been held back by the poor image quality provided by conventional surveillance systems. The problem with the proper functioning of sophisticated intelligent surveillance software is the old classic: ‘Garbage in - Garbage Out’. In other words, if you feed any image analysis system with the typically low quality images produced by traditional CCTV systems the software that enables some of the newest trends in the industry simply won’t function as advertised.
The result? False positives, missed criminal identification, no ability to produce reliable trigger events, and many other potentially damaging actions. One of the reasons that biometric identification systems have not been deployed in great numbers has to do with the quality of images that feed the systems and the resulting false positives.
In short with garbage in, intelligent surveillance systems -- full of promise -- become useless or worse than useless because they end up ‘catching’ innocent people, locking down a building by mistake, or missing the theft of a precious work of art from a public gallery, or missing a known perpetrator.
Conventional CCTV surveillance cameras are manufactured using sensors designed for the entertainment industry, and the NTSC standard dates back to 1953. In order to truly understand the requirements of surveillance in a security application, one has to have a clear understanding of imaging and forensic identification.
In order to have a true performance surveillance system, there is no other choice but to utilize the best capture electronics consisting of scientific-grade sensors and recording all the image data by using lossless compression. Given the heightened tensions that exist in today’s world, it is no longer acceptable to rely on standard surveillance systems that are based on technologies designed for the entertainment industry. It is critically important for our security professionals and law enforcement agencies to be able to make better identification of criminals that are caught on surveillance cameras.
Better quality images result in better identification, which in turn results in more guilty pleas. An increase in guilty pleas reduces the overall court costs required to prosecute offenders who would normally fight prosecution due to the poor image quality from standard surveillance systems. Performance surveillance systems offer significant long-term savings since their high quality images allow security professionals, law enforcement personnel and the legal system to actually do the work that we ask of them. Identify criminals, prosecute them and convict the guilty. The future is looking good.
Jean-Pierre(JP) Forest, CPP, is Director of Security Solutions, Avigilon (www.avigilon.com).
For more information, please send your e-mails to swm@infothe.com.
ⓒ2007 www.SecurityWorldMag.com. All rights reserved.
|
.jpg)
