Annotating Oblique Imagery

Introduction

In this weeks lab we will process some oblique imagery taking by DJI phantom 3.The flights imaged three different objects. The purpose of this lab is to get high quality 3D images. We will do this by annotated the UAS data in PIX4D. We will annotated 3 images and I will have an image without annotation for comparison. The images were collected by flying the UAS in a corkscrew patter capturing multiple angles.

Methods

First you need to upload the images to PIX4D. When you process the data make sure you select 3D models for the processing options template.

Next you need to run the initial processing. Once the initial processing is done you will begin to annotate. You need click on rayCloud and then select an individual image from the calibrated camera.

Then you will select annotate. From here you will point and click masking out everything except the 3D image you wish to process. This is a tedious and time consuming process which requires patients. You can hold the mouse down to mask more quickly. I found it is easiest to zoom out to full extent and start by imaging the borders and moving inward and a slow but steady linear movement. You will annotate a few images and then do step two of processing, giving you your completed image. It is best to pick images that show different angles. The more images you annotate the higher the quality will be of your final image.

Results

The images varied in quality depending on how many images you annotated. These are the results for the front end loader without annotations. Front End Loader Before. These are results after the annotations were completed. Front end Loader after annotations. There are some difference with the images but the differences are not profound. Here are the annotated images for the Shed and Tundra.
Annotated Shed and Annotated Tundra

Discussion

I noticed little difference in my annotated images. I annotated 5 to 10 images for both and tried to get variation in angles. To yield better results you would need to annotated more images. I think to get the best results all images would need to be annotated. I noticed that the hydraulic hoses on the front end loader contained  lots of background within them. To get rid of this you would have to zoom in very close to select the pixels to be annotated. I noticed the bottom of bot the front end loader and tundra lacked data which hurt the quality of the final image. The flight path of the UAS would have to be lower to collect data for these parts. I think this would clean up the image greatly. The whole process of annotating is very time consuming. If this needed to be done for your company, I think this would be a good job to out source otherwise all your time will be annotated verses discussing results.