AIY-vision-kit-DRONE

In this project, I studied and experimented the capabilities of the AIY vision kit of Google for AI application embedded in drone. This small shield provides huge image processing capacity thanks to its Movidius unit. In addition to its small size and weight it is a very viable solution of AI on edge for small UAV both Cloud or Onboard application:

A complet description of my work is available HERE

I am currently working on 2 usecases:

1] Crack detection

Implementing an on-board crack detection-classification system, and integrate the solution into a small U.A.V. Current state:

• Crack detection classifier
• U.A.V implementaion

I used this data for training a MobileNet V1 (160X160, 0.5) to classify crack image. On this youtube linkyou can see a video of the implemented solution.

2] Cloud facial recognition

Facial recogniton use heavy computional algorithm and currently solve by two solution:

• full on-board system: using powerfull computer like Jetson TX2 (quite big and expensive)
• full cloud: cheaper but may bring latency

My idea is to use AIY kit to divide the task, I achieved face detection on board, and sent the cropped face to the cloud for recognition.

In this case, we save bandwidtch and minimize the latency but still using a small board such as Raspberry Zero. You can see a demonstration on [this youtube link](https://youtu.be/c9JQTAcH2Pc).

Current state:

• Face detection - Cropped face solution -Cloud
• Face recogntion using facenet or OpenCV
• U.A.V implementaion and face tracking