No problem. BTW, just consider it as a brainstorm. I just let my thoughts flow freely.

The challenge for your problem is: does the satellite image large enough to see the lane divider.

Image for post
Image for post

Let’s brainstorm with the image above. First, you need to train your object detection with samples include arrows and the white lines. Even YOLO9000 does not train with those objects, so you need to collect those training samples with the proper labels. Second, you have to make sure the detection algorithm does not scale down too much internally so those objects cannot be detected anymore. Then come into the orientation of the object. I personally think that is not a critical issue. Once you have some kind of boundary box, you can add one or two dense layers to calculate the orientation if your training data have those labels. With data augmentation, create objects in different orientation should not be hard. Whether it is Mask-CNN or it is aligned vertically does not seems the hardest problem to solve. Collect enough training samples may be the hardest. You will for sure need to tackle many problems since your objective does not fall into typical object detection problems. But maybe identify the hardest one and solve them first.

If the lane is hard to see, there is another paper that I can recall using GAN.

https://arxiv.org/pdf/1611.07004.pdf. But the result can be varied because training GAN is hard and the accuracy is not too encouraging sometimes also.

Image for post
Image for post

Deep Learning

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store