Full disclosure, I’m the CEO of Solink, a VSaaS solution that is purpose-built for businesses through the integration of video and data (POS, Access control, etc. - solink.com/partners). Solink is both a modern VSaaS and a Business Intelligence (BI) solution in one. I was compelled to highlight an opportunity we see taking shape in our industry within AI & Video Analytics.
AI is made up of 2 distinct areas - namely, Training and Inference. Most inference models are trained on static datasets (large swaths of images that act as labeled “truth data''). Inference models are then applied on cameras of varying resolutions, field of views, environments to produce an output. You can often tune the model to produce different outputs through configuration parameters that are sometimes exposed or hard-coded in the model. The more images used during training, the more environments that can be accounted for, but it also makes for more configuration parameters and more computationally expensive hardware (GPU/Memory/etc). We believe the future of AI in video surveillance is one where each camera will dynamically and autonomously train itself based on what it’s seeing rather than static training. You will likely have a “learning phase” where the system is configuring itself. This will lead to smaller compute form-factors because there’s no need for the model to account for data that isn’t relevant.
**** *** **** ********?
****’* *** ******* ** *** ** are ******** **** ** ******
********* ** ******/******* **** ****** ** camera ** *** *** ** *** easy. ***** ***** *** ****** ******* are, ** *******, ******* & ****-********* to *********. ***** ** *** ******** many ***** ** ***** ********, ** believe *** *** **** *****-***** ************* for ** ** ** *********** ************* requirements ** ******* *** ***’.
**** ****, ** ********* * *** feature ** *********** "****-***" ******** *******. This ****** * ****, **** ** configuration ********, ** ****** * ******* as **** **** **** ****** ** camera ** **** ** ***** *********. Solink ******** **** **** **** ****** per ****** *** *** (********, *******, changes ** **********, ***). ** *** this **** ** ***** ** ** model ** ***** * ****-******* ** links ******* ******* ***** ** ********** activity.
** *** ****-******* *****, **** *** represents * ******; *** ***** ********** the ******* *** *** ** ******** links. *******, ** *** **** ** map ******** ******* (*-* ***********) ****** a ****** ** ******** *** ******** between ******* ** *** ** **** in *** **** *****.
** ******* **** ** * *** step ** ******** * ********** ** connected ******* ***** ** **** ******** than * ****-******* *****-****. **** **** allow ** *** ******** ********** **** as *****-****** ******** *** ***** ******* in * ********** **** (********* ******) rather **** ********** ******* (***** ** Google ****** ****, ** * ********* example)
******* ****… ** **, *****, ************ or **** ** ***** ** ** interest ** *** ** **** ********* we’d **** *** *********** ** ****, feel **** ** ***** *** ** me ********@******.***. *******, ** *** *** ********** in ********* * ****** ******, ** are ****** ** **** *********** - solink.com/careers
** ******* ******** ** **** ****** like **** *** ****** ** ** ignored ** **** ***** **** *****. I **** *** **** ***** **** over ******.
**** ****:
* ** ***** ****'* ***********. *** might *** ** ***** *** ***** was * ******* * *** ***** ago **** ******* **** -**** ************ (*******). ***** ******* *** **** ***** company *** **** ******* ** ** was ***** ********* / ******* ** do **.
******** ** ** ** * ******* is * **** ****. *** ********** cost *** **** **?