Axis on Video AnalyticsBy: John Honovich, Published on Feb 28, 2009
Axis is advocating using analytics on IP cameras. This is not news but I was surprised to hear their questionable rationale (see the 3 minute video). Axis claims the main problem for video analytics is that the price is too high for running analytics on central servers. To the contrary, the industry consensus is that analytics' main problem is poor performance - whether on servers or cameras. Indeed, performance historically was better on servers simply because of far greater processing power at far lower cost. Poor performance is a key risk of using analytics on existing IP cameras. The most accurate analytics in the market all use either purpose built cameras (like ioimage or sightlogix) or they use specialized chips optimized and dedicated to video analytics (like ObjectVideo on TI chips). The problem with Axis' approach is that running analytics on regular IP cameras greatly increases the risk of performance problems. Axis often cites via:sys and AgentVi as two examples of analytics that run on their cameras. However, my tests of via:sys show poor performance. AgentVi tends to work better but, of course, still requires a central server. Also, when loading analytics onto existing IP cameras, you need to be very careful about processor overload. You can easily cause performance problems by simultaneously using advanced compressions, high performance, multi-stream and video analytics at the same time. In this interview, Axis claims that with IP cameras you can "take any analytic from any analytic vendor." This is only true conceptually. The practical issues I outline make this very unlikely with today's Axis cameras. Quite frankly, I am dissapointed with Axis' stance. One of the things that I respect Axis about highly is their commitment to rolling out products that work really well and about the care they take in not endorsing science projects. Unfortunately, this interview makes assumptions that are technically quite questionable.