By Jon Cropley
Principal Analyst
IMS Research
Video content analysis (VCA) software can be run on standard off-the-shelf computers or embedded in video surveillance devices such as network cameras and encoders. In server-based solutions, the analysis is usually done centrally with the full video stream sent across the network. In embedded solutions, the VCA software is loaded on a digital signal processor (DSP) or embedded processor which is physically installed in the video surveillance device at the time of manufacture. In some cases, the software is up-loaded directly into the video surveillance device as a plug-in, whilst the device is in the field. Embedded solutions are often referred to as “analytics at the edge” as they have the ability to analyze the video before sending it across the network.
Server-based solutions offer more processing power than can be provided at the edge. For this reason, processor intensive applications, like face recognition, are normally server-based, whilst edge-based solutions are restricted to simpler applications, such as people counting and tripwire. Typically, the vast majority of the processing capability of the main processor is needed for system control and image processing tasks, leaving limited resource for VCA. Whilst it is possible to add a dedicated analytics processor, this increases the bill of materials and can make the price of the device uncompetitive. However, IMS Research predicts that in the coming years the types of applications that can be performed at the edge on the device’s main processor will increase. This will mainly be driven by the availability of more powerful processors and partly by the refinement of VCA applications to make them less processor intensive.
As powerful processors at affordable price points become available to video surveillance manufacturers, they will increasingly add VCA to their devices. Basic analytics will become standard features and more advanced “paid for” analytics will be performed at the edge. Over time, an increasing variety of analytics will be added as standard features, at no extra cost to the customer. Moreover, more powerful processors will enable multiple VCA algorithms to be performed at the same time, for example loitering and object tracking to improve the detection capabilities. Another possibility is cross camera tracking, where one camera hands-over to another to follow a person as they move through a building.
If Moore’s Law continues to hold then the power of processors will continue to increase quickly. Moreover, VCA is increasingly being used outside of the security industry, such as driver assistance cameras in cars and interactive gaming console peripherals (e.g. Microsoft’s Kinect). Some of these new markets for VCA offer high volume potential, which is attracting the attention of the chip makers. In the coming years it is likely that these companies will introduce powerful processors that are optimized for VCA applications. For the security industry, this will inevitably mean more analytics at the edge.
About the Author
Jon Cropley is a principal analyst in IMS Research's video surveillance and VCA research group. Before this time he was the director of its Automotive & Transport group. Jon joined IMS Research in 2001 and is a highly experienced analyst, having authoured numerous syndicated research reports. Jon worked for TRW Automotive in Germany before joining IMS Research and holds a BSc from Lancaster University. He is based in IMS Research’s headquarters in Wellingborough, UK and may be contacted at [email protected].
