Processing Updates and GIS Day Template

Processing Updates

This week, we processed one of our Purdue Wildlife Area flights in Pix4D. The purpose of processing an orthomosaic of this flight is to compare the time it takes, and the product quality we can get from Pix4D to our Loc8 searches.

 Pix4D took an incredibly long amount of time to process this flight. Pix4D recommends having an overlap of at least 75%, and a sidelap of 60% for general use cases of orthomosaics. We flew this specific mission with 70% overlap, and 70% sidelap. Slightly less than recommended for overlap, and significantly higher than recommended for sidelap. The DJI Mavic 2 that we used was actually unable to finish the entire flight in one battery, seeing that a mission with 70/70 requires many flight lines. In fact, 70/70 requires more than three times the number of waypoints as a 10/10 flight does. To get into our processing though, figure 1 and 2 below show the quality report given to us by Pix4D.

Figure 1: Pix4D Quality Report generated from orthomosaic processing

Figure 2: Pix4D Quality Report generated from point cloud and mesh processing

As can be seen in the figures above, Pix4D took an incredible amount of time to process the imagery. Just the initial processing took over three hours to process. Combine that with 1hr 40min for the point cloud, and another 34min for the meshes, we are looking at over 5.5 hours. Without the point cloud and meshes, the orthomosaic was scarce at best. Enormous chunks were missing, and you could not possibly locate the target. Add in the point cloud and meshes, and decent imagery is seen. Figures 3 through 6 below show results of the processing.

Figure 3: Full orthomosaic image with point cloud and meshes enabled

There are many chunks missing in the full orthomosaic. This could be in part because of our slightly lacking overlap, or because the images from those sections were not clear enough to successfully stitch together. Either way, this is a glaring problem with Pix4D. Sometimes, there's just issues with the processing. Thankfully, all the targets we were looking for do show up in the imagery. However, the massive missing chunks could be the difference between finding the target and not.

Figure 4: Neon yellow shirt found

Figure 5: Red shirt found

Figure 6: Turquoise shirt found

Figures 4-6 above show all of our targets. Each one of them was found, which is great. However, the time it took to process the imagery in order to find them is a huge problem. Additionally, once the imagery is done processing, we have to manually look through the orthomosaic in order to actually find the targets. That part of the process is barely any better than simply looking through the raw images to find the targets. At least that method, squinting, has shown moderately comparable results thus far. Pix4D, however, does have other issues.

Figure 7: Pix4D orthomosaic terrain level generation issue

 Figure 7 above shows another huge issue with processing imagery in Pix4D. Even though in the full orthomosaic, from a top down view, this doesn't seem to be an issue, having terrain leveling issues like this is a major problem. The ground at PWA is very close to completely level, there is certainly not the enormous spike in altitude at any one area, especially not to the extent that Pix4D shows. This could potentially be fixed by adding ground control points, and having better RTK or PPK on the platform. Once again, though, this is an unnecessary time sink. Pix4D already took over 5 hours to process the imagery, plus longer flight times by about 20 minutes. Any other added time to get better imagery is completely wasted, and could be very valuable time lost. In the amount of time Pix4D took to process imagery, then locate our targets, Loc8 could have done the same thing without even leaving the field, and in about 5 hours less time.

Discussion

Our main goal of this research is to prove why processing data in Loc8 for search and rescue is a much more efficient method of SAR procedures than anything else. If nothing else, we have proven that Pix4D is absolutely not an acceptable method of SAR. Between the time Pix4D takes, to the product it gives us with missing chunks and terrain level issues, to having to manually look through the orthomosaic to find the target after everything else, it's just simply too time consuming. Loc8 is able to process the same amount of images in more than 5 hours less time, while also pointing out the target for us. We will continue to experiment with Loc8, and find the best method of using the software for SAR.

Poster Template

One of the products we will generate for one of our final deliverables is a poster. This poster will highlight key aspects of our research, such as our objectives, methods, and a variety of our results. This poster will be our centerpiece for Purdue University's GIS day in early November. Right now our poster is only in template form, but as our research comes together, the poster will be populated. You can expect to see the final draft of our poster here as a separate post as it becomes available later this semester. A link to our poster template can be found here.