Vector_AI_Vision

Bringing advanced computer vision on the Anki Vector robot. By using Vector's API to fetch image data and TinyYoloV3 neural network, vector is able to recognize objects in a scene.

How does it works ?

• Vector's API are used to fectch image data from Vector's camera
• Images are preprocessed (reshaped,...)
• Preprocessed images are sent to TinyYoloV3, which is a smaller version of Yolo optimized to reduced computational needs (which means higher framerates for less computational power)
• Output of Yolo are decoded
• Results are printed on Vector's screen + the captured frame from the camera
• Results can also be spoken by Vector (which is cool in a first place but annoying finally)
• And the process goes over and over again

Technical needs

• A Vector robot here: https://anki.com/en-us/vector.html or here https://www.digitaldreamlabs.com (the second link is the new owner of the Vector)
• Tested on Python3.7 with dependencies:
  • tensorflow
  • numpy
  • Pillow
  • anki_vector SDK from: https://github.com/anki/vector-python-sdk
  • (It should be all the dependencies, sorry if I missed one)

References

• Official Anki's Vector SDK: https://github.com/anki/vector-python-sdk
• Yolo models: https://pjreddie.com/darknet/yolo/