How would you qualify the "excellent results" statement with a percentage? I suspect that a choke point with high 2-way traffic will be roughly 90-94% accurate if that. How does the counter analytic handle groups close together or with overlap? Is an overlapped group a single count? What about a person pushing a shopping cart of baby stroller, is this a single entry or is a shopping cart counted as a person? What about children? Lastly, how was accuracy in this situation verified? Was recorded video reviewed, people counted by a human and then comparisons done for accuracy measurement?
I am based in Wellington and have been asked by a client to look at people counting in their 6 stores (5 Wellington 1 Auckland). Presently have 3 stores with Hikvision IP CCTV and looking to upgrade the others in time. Which supplier do you work with as I currently use Atlas Gentech
#2, we have no plans to test Mobotix's people counting offering. The main issue is that Mobotix has limited market presence. That's not a judgement of their product quality but a reflection of the limited interest in Mobotix products for our overall membership.
Our next people counting test, if or when we do, would most likely be Axis with Cognimatics given Axis broad market presence.
John, I understand that Mobotix is not a 'mainstream' offering like Axis or Hik, however, if more of us Mobotix users choose not to let you know that we would like to see them tested more frequently by IPVM, it is unlikely that you ever will. Additionally, if your testing prooves to be favourable it is likely their use will become more 'mainstream' due to their outstanding performance and flexibility. I find that most of their analytics offerings are brilliant and would like to see them compared to others in the market like Bosch and Axis (Cognimatics).
Ted, sure, please let us know. We definitely take comments / requests into consideration and we have been tracking Mobotix opening up since the Founder's departure. For example, if Mobotix people counting is integrated into some mainstream VMSes that would increase the likelihood of testing them.
For example, the downside of Hikvision's people counting is its lack of 3rd party VMS integrations but since Hikvision is so widely used, it still is of broad interest.
This is not an accurate way to test people counting. Single person walking. There is no change in ambient light. The ideal way to stress test a counting system in retail that all retailer analytics vendors like Brickstream (FLIR), Xovis, Delopt, Irisys do is count 100 people in and 100 out within 2 mins. Infact in most Malls and retail, this is an acceptance criteria. Check out Dubai Mall, they have close to 1000 sensors in their mall. On the fountain entrance they have close to a flow of 300 people every minute. Large amount of the installs that I have been involved with required 99.5% accuracy with counting, I know for sure 2-D camera systems fail.
The 4 companies share a market share of close to 80% of the global counting manufacturer market.
Installation such as this is not the best way to test. 2D camera systems in real world installs have only a 90% accuracy at the best. Seen and validated over 20 different vendors. The three companies mentioned above are the best. Delopt I know can count in 100% darkness with the same 99.5% accuracy. All the above companies, at the same time, detect adults, kids, groups, manage u-turns.
Side Installs. How was your parallax test? How did it manage short people being obscured.
Yeshaswi, thanks, informative first comment on IPVM!
I agree with you about high volume impact and added a note in the findings:
NOTE: Our test was done in a low to medium traffic office environment and is only applicable to similar scenarios. By contrast, heavy traffic retail stores are likely to have worst results, though how much worse we can not say because we have not tested it.
One thing I do think is worth noting is relative to this comment:
Check out Dubai Mall, they have close to 1000 sensors in their mall. On the fountain entrance they have close to a flow of 300 people every minute
Most American retailer stores are lucky to get 300 people in a half a day. That might be an exaggeration but only a slight one.
And Hikvision, at least in America, sells best in smaller stores and offices with low traffic where people counting is likely to work well.
Your point remains - the rate of traffic will have an impact especially if there are numerous people continuously entering in and out.