Actions
Engagement
  • Author Author: rscasny
  • Date Created: Date Created
  • Last Updated Last Updated:
  • Views 1691 views
  • Likes 1 like
  • Comments 15 comments
Related
Recommended

Poll: Seeking Your Interest in Sound+Vision AI Workshop: Demo on Detecting Sneezing or Dog barking?

A few days ago I spoke with a supplier is interested in conducting an AI workshop on Sound Classification and Vision Object Detection. This workshop sounds interesting because it combines two senses, vision and sound. I ran a poll on this workshop here: Poll: Seeking Your Interest in Sound Classification and Vision Object Detection WorkShop

 

This workshop demonstrates how today's products are becoming more sophisticated. It brings together vision and sound AI models that when used together enable more intelligent products making better application decisions. And it takes you through the network development flow for sound classification (detecting certain sounds) and vision object detection and walking you through our reference design that has sound and vision sensors sharing meta data that enable a higher level application to leverage certain events it sees and hears.

 

The supplier has two demo options:

 

1. Detect sneeze….then face detect…then detect is person wearing mask or not

2. Detect dog barking….then vision to confirm dog or not

 

Poll: which option would you prefer to see demonstrated?

Parents Comment
  • in reply to spinden

    Spinden,

     

    Birds should be possible although we get a lot of background noise from planes, automobiles, wind in the trees, shouting people and so on.  Especially as most audio amplifiers use automatic gain controls which just amplifies the background 'hiss/noise' when there is no target noise available.

     

    Dubbie

Children
  • in reply to dubbie

    One issue related to this may be with the birds that have started to mimic urban human noise pollution. e.g. you can now get birds mimicking car alarms. So the training data set may have to be tweaked to be able to accommodate this to prevent ending up with photos of 'the lesser spotted Ferrari'. Although in this case, just got to get the second stage vision classification process to successfully detect between a bird and a car in order to reduce the error.

     

    Another issue may be the regional localisation of birds. Existing data sets are probably pretty good for the region the primary research was conducted in but perhaps not so good globally.

  • in reply to beacon_dave

    imageimage   While this makes good sense, I can't help but chuckle thinking that my local birds have an accent.  Tweet, don't ya know, eh? Oh. tweet, you betcha.  (Wisconsin, USA)

  • in reply to dubbie

    Maybe the system could also try and detect a match against a database of known interference signals such as a car, lawn mower, children playing and positive matches nulled from the input signal. Such data could be collected in a highly urbanised environment to exclude birds singing or perhaps at night (maybe not the lawnmower ! ). Unfortunately my short overview would require a lot of processing power and coding to realise.

     

    You could perhaps filter on bandwidth to negate rumblings of tyres on asphalt etc or notch out some really specific interference like a neighbour's hedge trimmer.