Imaging Experts Help Me
I am a hobbyist and want to implement some advanced analytics implemented by an IP Camera
In Fact i am seriously in process of developement of my own IP Camera
i am stuck at selection of processor, i am unable to select processor because of the analytics which are need to performed are not that know to me
i am new to machine learning and deep learning or opencv
few the tasks listed below
Set of Analytics : 1. region entrance -a. human enters, b. vehicle enters, region A to B / B to A.
loitering - human staying for long time
tripwire crossover - same as region entrance
fence trespass - draw a line and some one crosses
unattended baggage -not human but object newly brought to view but not been removed more than certain time
stopped vehicle - a zone is drawn in which a vehicle is stopped for more than a time
object removal.- a zone is drawn from which if a object is removed then it alarms
What kind of perquisites are needed for this kind of analytics
Kindly help
Thanks
I don't understand your question - do you ask about hardware / software / skill prequisites? And you got a lot of things to do ^^. This is like a very very very big commercial software on top lvl technology, sigh.
Let me comment on 5)
The first thing you need to do is to locate the object. This is called object detection. For complex objects(like different vehicle types) you most likely want to use a neural network. Good thing is - there are some pretrained models for this purpose(car detection)- so you dont necessarily need to collect data and train.
Now here comes the complicated part "in which a vehicle is stopped for more than a time". This implies you want to identify(recognize) a certain vehicle to know its "exactly the same" and this particular car has stopped several times
In this case you need to use tracking or embeddings. Funny stuff. My advice - break things down and start small. It will be complex enough.
I mean to ask hardware prequisites which are decided based on software functionality requirements, so i needed some info which hardware would be prefereable
Is this intended to run on mobile / embedded device? If so - you can speed the detection with using some special kind of hardware you plugin into the device - usually via usb - check google coral / intel movideos / nividea jetson
My machine for this kind of stuff is a cloud machine with 4 cores / 16 gb of ram / nivdiea Tesla K80 gpu and 80GB of ssd diskspace.
The things is - its hard to give you any spec without knowing how many (concorrent) user you will have and what is acceptable timings for you. Will all the functionality listed above be on one machine(wouldnt do this - its scales bad and can have side effects)? Sorry this is too broad i am afraid.
Maybe also go for the cloud - its expensive but it usually scales pretty good.