Object detection using google's vision API (label detection)

Hi community,

I’m wondering if anyone has already connected a Mobotix cam to google’s vision API in order to use the “label detection”, i.e. identifying objects in an image.
According to google the label detection can “detect broad sets of categories within an image, ranging from modes of transportation to animals”.

The use case I have in mind:
When an event occurs, a picture (or the whole recording) of the scene is sent to the cloud to conduct the label/object detection.
If the reason for the event is an animal (e.g. cat, fox, hedgehog, bird, dog, …) then nothing to worry about.
But if there is a person or vehicle in the scene then send an alert as there might be a villain on the property.

Future development:
I hope we will see such a feature directly integrated into the Mobotix cams maybe as “MxActivitySensor 3.0”. Probably it can be achieved without cloud using pretrained models on the second core of the Mx6 models or together with special hardware as for example used for the AIY Vision Kit (large-scale website, the function I mean is located at point 111. object detection demo).
At least today it is neither cost-intensive nor computationally intense anymore to analyze a picture’s content.

@AdrianS If Mobotix does not have such a feature on the screen yet, please put it on the requested feature list.

PS: To get a feeling what is already possible today, just take an image file and drag’n’drop it on the “Try the API” box on the website of google’s vision API. The result will amaze you :astonished:.

PPS: Of course not only google offers object detection, there are others as well: Amazon Rekognition , Microsoft Azure Cognitive Services, …

1 Like