Milestone Machine Learning Camera Auto-Setting Examined

Published Mar 13, 2019 14:45 PM

Milestone wants to improve image quality using Machine Learning to solve the problem of "a camera doesn't know what it is being used for", automatically, without user input.

We spoke to Milestone, and in this note, we examine the offering, analyze its value, risks and industry impact.

MIPS ************ ********

*** ********* *********** *** ********* ** MIPS ****, ****** *** *** ** machine / **** ******** ********, **** a ***** *************. *** ********* **-****** presentation **** **** ******** *** ******** goals:

No ********** *** ************ ** ********

********* *** *** ****** ** * timeframe *** **** ******* ***** ***** to ********, ** **** **** ** would **** ** *****. *******, **** noted **** *** *********** ******** *** be ***** **** ********** ****** ** months, *** *** *********** *&* ***** working ** **** ******* ** ********* assigned **** ******* ** ******** ** innovations **** *** *-* ***** **** release.

3 *****-**** **********

**. ******, *********'* ******** ******** [**** no ****** *********] ********* *** * "Missions" ********* ***** **********, *** ********* camera *************:

  • *****-******** **** ***********
  • ******* ***** ***********
  • ******* ******** *****/***-***** ************/********** ****

LPR ********** *************

********* **** * ***** *** *** to *********** *** ************ ******** *** camera **** ***-** ** **-*****:

***** **** **** * ********** *** car, **** *** *** *********** ********** the * ******* ********** **** ********* license ******, ***** *** ******* ******** from **********. ***** *** ******** ** blurred ******* ****** **** ***** ****** the ************, *******, **** *** *** note *** ****** **** *** *** benchmark ******-***** ******* ****** *** ******* changes **** ** *********'* ****-********.

Server ***** ********* ********

********* **** **** *** ***** ******-***** deep ******** ** ************* ****** * camera's ****, *******, ****, *** ******** settings. *** *********** ** *********** *** could ** *** ** * ********* recorder ** ****** ******.

Latency ********

*********** ****** ********** ***** ** ** issue *** ******* **** **** ** quickly ****** ******** ***** ** ******** lighting *** ****** **********. ******* ** a ****** ** **** ** ****** with *** ********* ******* ** ******* video **** *** ****** ** * server, ****** *** ****** ******* *** video *** **** **** **** ********. This **** ******** *** **** ** its *** ******* ********, ** *** scene *** ****** ***** ****** *** commands *** ********* ** *** ******.

Camera ****-******** ******* *****

******* ******* *** ******* ******** ** optimize *** ****** ******* *** ******* parameters (**** ** *******, ****, ***** balance, ***.). *** ** *** **** Milestone ******** ****, ** **** **, cameras *** ****** ** **** ******* advantage ** ***** *** ********** ********** resources *** ******** ** ******* ********. Plus, **** **** *** ********* ** knowing **** ********** *** ****** *** camera *** *** ** **** **** them *** ***** **********, ***** *********, by **********, ***** ********* ***.

Deep ******** ***** ***********

********* **** *** ********* *********** ***** ~40 ******* *** **** *** ****** and ******** ****** *** ** ******* the ****** *******. **** ** **** for ******* ***** ******** *** *************. Because ** **** *********** ******** **** (time *** **********) ********* **** **** are ******* ** ********** **** *** the ********* ******, ** * "******** service" ** ******.

**** * ******** *********** ** ***, they **** **-*********** *** ** *** at *** **** ********* *** ** much ******* **** *******.

Prototype ******** **** *** *****

*** ********* ** *** **** ************, Milestone **** **** ***** ********* ** only ********** **** "**-******" **** *** Canon *******.

*******, **** ***** *** **** ********* would ******* ***** *** ******** ****** driver ** *** ******* *** *** party *******.

*******

*********'* *********** ** ***** *** *&* and ********* ********* **** *** *** actually ******* (**************) ***** ******** ***** ********** **** the ********** ** *** ******** **** take ***** ** ******* ** *** never *********.

** ****** ********* ** *** ***** of **** ******** ***** *** ********** of ***** **** ******** ** **-***** the ******, ***** *** ********** **********. However, ** ********* **** **** ******* to ********, *** ** ***** ************, it **** ** * ***** **************.

Vote / ****

Comments (8)
UM
Undisclosed Manufacturer #1
Mar 14, 2019

This technology will already be present in the hardware by then. Camera manufacturers don't want their image quality being dependent on a 3rd party VMS, especially a sister company to a major competitor. I know of at least one company that can change settings automatically based on dynamic light conditions or on a schedule. If it is something very specific like LPR, there are built in templates that adjust shutter, gain, and color for best practice.

(3)
(1)
Avatar
Igor Falomkin
Mar 15, 2019
AxxonSoft

Absolutely agree. Especially taking into account that fact that all SoC vendors for cameras are adding hardware acceleration of neural networks on board of their chips.

(1)
(2)
U
Undisclosed #2
Mar 14, 2019

I beg to differ, Milestone's "willingness to spend and R&D and marketing solutions they are not actually selling" is evidence that shows they have a technology road map, a direction and a goal to reach.
there is nothing wrong with announcing what you want to do in the next 3-5 years, intel for example does it too. 

i would be happy if the other companies in the industry would do the same, it would be easier and more interesting to know what to expect.

 

(4)
(1)
JH
John Honovich
Mar 14, 2019
IPVM

#2, thanks, that is an interesting counter.

Intel makes sense on the chip side since those chips need to be incorporated into other products by other companies, so the others need time to plan out. Same reason why Ambarella in our industry gives longer-term guidance / outlook on what chips are coming.

To the contrary, product manufacturers generally do not. Whether Axis or Apple or Avigilon or whomever, they generally only announce when the products are ready or just a few months away. And when they do, like Hikvision on AI, the gap between announcement and actual shipping working products reduces confidence in buyers and even in some of the company's own employees.

The in-between approach is manufacturers privately previewing the roadmap with individual partners. This allows a limited number of people to be aware and start planning. To the contrary, getting on stage in front of 700 people and livecasting it on YouTube is a whole different approach.

U
Undisclosed #3
Mar 14, 2019
IPVMU Certified

Not mentioned in the MIPS presentation, Milestone told IPVM their prototype is only compatible with "in-family" Axis and Canon cameras.

However, they noted the next prototype would utilize their own XProtect camera driver to add support for 3rd party cameras.

this is odd as it makes it seems as though XProtect has a native mode which is Axis/Canon compatible, and that other cameras are accessed thru their 3rd party drivers.  from what I’ve seen though, Axis cameras use “their own XProtect” drivers just the same.  

maybe they’ve had to enhance these drivers to support some of this new functionality?  if so, it might be a while until all drivers are at that level.

(1)
Avatar
Sergio Guzman
Mar 14, 2019
Pine Crest School • IPVMU Certified

I doubt it will work initially, but they need to start the project and have it hit the dynamic complications of the world for field testing. I welcome this attempt. This is no different than the IPVM calculator. Is it perfect? No. Does it mean it's still not awesome and useful? Nope. It's still useful.

(1)
(1)
PG
Pavel Grozdov
Mar 14, 2019

Whilst I don’t think it will work perfectly out of the box, I think it’s a step in the right direction.

I think it’s going to be a case of buy the expensive camera with it built in or buy the cheaper camera and have Milestone be the brains.

In my experience having it done centrally is better as it can be updated and bugs fixed very easily, one place to manage it all, universal interface etc

JH
John Honovich
Mar 15, 2019
IPVM

I think it’s going to be a case of buy the expensive camera with it built in or buy the cheaper camera and have Milestone be the brains.

Pavel, that's a reasonable possibility.

One challenge I don't understand is that presumably their sister company, Axis, who is the most widely used camera manufacturer on Milestone VMSes, is going to develop this themselves (no inside knowledge - just obvious tech trends). If only Axis would go on stage in front of 700 people and tell them what might be coming in 3 years!

(1)