Video content analysis (VCA) has garnered a tremendous amount of attention in the past few months and, if all is to be believed, the technology is near the point where it is a “buy” rather than a “sell.” And while this is good news in our beleaguered economy, there still remains some question as to where the VCA should be located – on the edge device (camera) or on the head end – or on both?
Before addressing the options however, a definition of VCA is in order. Video analytics is a process that extracts useful information, in the form of data, from live or recorded video images. Applications include the capability to identify loiterers, detect vandalism or monitor crowds for abandoned baggage. The capacity to accomplish this, however, depends on where the VCA is located.
The Case for Edge-Based AnalyticsIn a recent report on the market for VCA, IMS Research forecasts that its penetration into network surveillance cameras will exceed 40 percent by 2012 and it goes on to predict that network cameras will be the main hardware platform for embedded analytics. The advantage of edge-based embedded analytics is bandwidth preservation and reduction of storage requirements because the need to transmit all captured video for analysis is eliminated. In addition, content analysis on the video of interest can be performed when the video is in its highest quality and before it is compressed for transmission over the network to a recording/storage device.
Improvements in imaging technology, including digital signal processing (DSP), megapixel technology and increased sensitivity to low light have also contributed to the increased adoption of camera-based analytics. DSP chips deliver more processing capacity, which enables more edge processing; and megapixel sensors offer greater resolution and detail, which are necessary for analytic products to perform complex operations. Intelligent cameras can be programmed to transmit at low-resolution rates until a pre-defined event of interest occurs, at which point transmission switches to a higher frame rate and resolution. Improvements in compression technologies such as H.264 have also enabled transmission of higher definition resolution over a lower bandwidth.
Cost is another factor influencing the move to camera-based analytics because it is a more effective way of implementing VCA. Single unit cameras featuring built-in analytics can be installed where and when required, or the number added to as required. And when more processing is done at the edge, it can help to reduce the cost of analytic processing by eliminating or downsizing server requirements.
Efficiencies at the Head EndDue to the large installed base of analog cameras, however, IMS Research predicts that in the short term the biggest penetration of VCA will occur within network video recorders (NVRs), digital video recorders (DVRs) and video encoders, which convert analog signals to network compatible signals. It is argued that relying solely on camera-embedded analytics is not as efficient when analytics must be performed or metadata correlated on enterprise-wide systems such as casino or educational installations, where hundreds of cameras are deployed. Additionally, forensic searches (i.e. facial recognition, object identification, etc.), which are usually performed on recorded video, are still beyond the capabilities of most on-camera solutions at this time.
In these instances, the greater processing power and central management capabilities of a server-based or head end solution is generally a better solution. Centralized analytics allows for the configuration of different application sets on different cameras and at differing times. For instance, in a car dealership lot, the parameters for motion detection would generally be programmed differently for nighttime recording versus daytime and this is done more conveniently through centralized management. In another example, VCA may be used to calculate periods of peak activity or for counting numbers of people passing through an entrance and this function can more easily be achieved with a head end solution.
Whether it’s camera-based or server-based, VCA provides benefits to the user in terms of data mining and retrieval speed. Data mining of POS transactions for instance has been used extensively in the retail industry for creating marketing promotions, identifying buying trends and developing new product. On the security side, exception reports that include surveillance video and accompanying POS data can be generated, and as an example if the POS cash drawer is open longer than two minutes, the system will create an exception report with the time, date, employee, cash register, and revenue center along with the video of the event.
VCA’s FutureSo, back to the original question: Is it better for the VCA software to be located on the edge device, the head end (i.e. server), or on both? Or is there yet another answer, such as imaging processing over IP (IPoIP) technology, which distributes the image processing responsibilities between the camera and the server?
Whatever the choice, VCA is another maturation point of our industry and one from which much of the future of the industry will developed.