Using Loc8 to find a missing person

Introduction

Figure 1

Loc8 is a specialized program that allows a pilot to search through thousands of photos for a specific color range. After it has a "hit" in one of the photos it flags the photo for further, human, review. When reviewing it places a circle around an object that falls within your color spectrum with a large red circle, see figure 1. In this case figure, 1 is a truck that happened to have a blue-ish reflection around it. The program flagged the blue reflection and circled it in red. Next to its up to the person going through the data to determine what they want to do with the image that had a "hit". If the item circled was possibly what one is looking for then the person going through the data would flag the image. If it is something like this truck, which is not what we're after, then the image can be archived. For this trial run of Loc8, the "missing person" or target is wearing blue jeans and a black shirt. The software is very sensitive to color ranges and will probably come up with multiple false images that require a human eye to determine whether or not its the target, these are called "false positives"

Method

Figure 2

Figure 3

False positives are something that can waste a lot of time and cause fatigue when going through the data. This can cause data to be missed solely based on the sear number of images that one has to go through. Something that can increase this is having too big of a range or too many colors. This will have the software looking for a whole bunch of colors and flagging hundreds of photos.

This can be limited in settings, figure 2. By setting a minimum pixel per cluster you set a minimum amount of pixels that can be within the color range (figure 3). Much of the other settings weren't used in this example so that many data points will be detected and give the greatest chance of finding the missing person. By selecting the minimum pixels per cluster you set how many pixels are needed for a "hit" in the data. This can also work in another way by limiting the maximum pixels per cluster by limiting the number of pixels in a color range. This can be helpful if a color like black is being selected, as all shadows will then create "hits" in the images and make it very hard to see if there is the missing person actually in the photo. When setting a color range it can be important to remember what the colors of the area will be. If it's a lush summer day, green would be an important color to avoid, as grass and treetops will fill the possible sightings photos. When looking for a missing person it will be important to focus the program in on colors that aren't everywhere, this will limit the number of false positives and allow for an easier time when going over the results.

Discussion 

When looking through this set of data we were given images of the clothes that the "missing person" was wearing. This was to provide a color range and determined what it should be set at. Even with the color range that was given there were about 50 images that were "hits" and only 3 of them had the missing person within the photo. This is why going over data like this is so hard and programs like Loc8 are essential to narrow down the number of photos that one has to look over to find the missing person. As 50 photos are a lot easier to comb through than the 1000+ that was started with.

There are times when just looking through the data might be a quicker option. As stated earlier black can be a very hard color to scan through many images, this is when a manual scan through the images might prove better results. This can also be the case if many of the photos are of treetops where something could be underneath the trees and not even Loc8 would be able to detect them. The quickest way to then go through the data would be to go through the treetop images and just focus on looking into the clearings. These clearings are usually shadowed and already show up if the color selection is on black, but with all the red circles that can be contained inside an image, it can be easy to miss something. This is what happened when looking for extra credit with this assignment. When looking for the extra hard missing person in this assignment the image that contained the person was flagged but due to the sheer number of false hits on the shadows around where he is, and the circles that surround the "hits", it was impossible to see the person. After going through the data it was decided to go through the images by hand to check again for the missing person. After going through it and focusing on clearings in between trees the body of the dummy was found, which can be seen in figure 4.

Figure 4

Conclusion

Using drones in search and rescue (SAR) can drastically speed up finding the person. A drone can cover ground many times faster than if a search party is sent out, the only drawback is it can be less thorough. By using software that can detect when color is present in a data set the lack of thoroughness can be made up by highlighting images that could potentially have the missing person within. This can drastically speed up a SAR mission and quickly get the missing person found.