How it works: The two main processes are machine learning and computer vision. We gather a set of images or video frames and mark the target object in the image that you are trying to detect and label it. Then we run it through cloud processing and that will output a model. The model is then loaded onto an embedded computer and integrated with the on board payload.
Strengths – Simplifies training/setup, Can perform otherwise very difficult image processing tasks
Weaknesses – Only as good as your data set, Can produce false positives
The image to the right is of semantic segmentation of aerial imagery. Red is buildings, green is foliage, gray is pavement/dirt/etc, pink is vehicles.