Video of a Video Case Study

Nathan Rose:

Now, this was a video that came from a surveillance camera system, and it showed a bus pulling into an intersection, crossed the path of a motorcyclist. In the video, we were able to see the bus pulling out. We were able to see the motorcycle entering the frame of the video. Ultimately, the motorcycle goes out of the view of the camera, behind the bus and the collision occurs. This video gave us a lot of information about the crash.

Connor Smith:

Like most of our cases, we've had a few materials come in and we reviewed those. And then as we got more information, we figured out a plan to analyze the case and figure out what we needed to do when we went to our site inspection and what information we could glean from the documents that were provided to us.

Nathan Rose:

One of the interesting things about this case was that the video we were provided with was a video of a video. A crash will be captured on surveillance footage, and when the initial investigators go to get that footage, no one knows how to get it off of the surveillance camera system.

Connor Smith:

The video wasn't steady. You could see that the camera was moving, and so we could watch the video. And our first pass through there was just to figure out what we could learn just by watching the video.

Nathan Rose:

We go to do the video analysis and we establish positions for the motorcycle from the video. They have to be consistent with where on the road that physical evidence is deposited.

Connor Smith:

We can notice about halfway into the video once it enters the screen, and we saw some weight shifting towards the front wheel, so that to us was an indication that braking was applied. And then after we viewed that security camera footage, we reviewed the photographs that were taken by the police. In those scene photos, we could see where the vehicles came to rest, debris, and there's actually a tire mark that we located. And then once we got all that information in there, we decided we were going to do camera matching photogrammetry to try to locate that evidence on the road.

Michael Erickson:

We went out and we scanned the site and the photogrammetry part of it is how we tie the video and the scans together to create a virtual environment where we can get realtime measurements in real time speeds from the vehicles going through the scene.

Connor Smith:

We used the [inaudible 00:02:13] photographs to find all that evidence on the roadway, and then we also camera matched that security camera footage so that we had multiple vantage points of the scene, because all those things are basically happening at the same point in time, at the same point in space. Using those different methodologies allowed us to join them together during our analysis process.

Nathan Rose:

When we were doing a simulation of this, we had positions of the bus and positions of the motorcycle that were established from the video analysis. But one of the things that made this case really helpful is that it involved a Kawasaki motorcycle that had an event data recorder on it. We were able to get speed data for the motorcycle from that Kawasaki EDR that gave us something we could compare the video analysis to. We knew from the EDR data how fast the motorcycle was going at various points leading up to the collision.

Connor Smith:

That was one of the other important parts about getting that event data recorder for the Kawasaki, is that you took that data backwards, it took it off screen. Taking that information then extending it back in time can really help you to analyze and get the bigger picture of the crash instead of just this little section of the crash right when the crash occurred.

Nathan Rose:

On many of our cases, the way we're going about our work, the analysis and the animation are intimately connected. Really by the time we get to the end of it, producing the animation is really just rendering what we've already done for the analysis, and that's certainly the case in this instance.

Michael Erickson:

I've built a workflow around using the data points that we have so that we're not doing a lot of extra work when it comes time for animation. We're using the point clouds and we're using the models that we're doing the simulation with, but we have materials and lighting applied to them that are ready to go for animation at any point.

Connor Smith:

We can use our computer environment to analyze it from so many different perspectives, points of views. We can show what was the school bus driver doing when there's cars passing in front. He's trying to turn left into a stream of cars. We can validate or verify some of his observations from the crash, making sure that's all lining up. The best way to look at that is in our visualization software. As part of that, we just are naturally progressing to a place where we can make an animation that will help potential jurors, that will help them understand the incident from more than just the single video that captured the incident.

Nathan Rose:

There's a lot of literature in our industry that is talking about techniques for video analysis, but the thing is, none of that literature in the past has really addressed the accuracy of those techniques when we're dealing with a video of a video. That was a unique opportunity on this case for me and for establishing the reliability of our techniques for future cases to essentially validate that yes, those video analysis techniques were reliable. They agreed with the EDR data and that the conclusions we're reaching are accurate. That's important for our work that we know that our techniques are producing reliable results.