The Physics of Fake Videos

This composite image shows the basic idea of video analysis.

Was that awesome video real or fake? How could you tell? Well, from a physics perspective there are at least three methods to use in an analysis that will tell us if a video is fake.

Video Analysis Method 1: Unrealistic Trajectories

When you think of video analysis, you probably think of this method. The main point of a video analysis program is to get x and y data for some object in each frame of a video. If you know the frame rate, then you get both x and y-trajectories for that object. Simple, right?

But how do you get the position of the object? That’s a whole different blog post, but maybe this will get you started. Here are a few more resources for video analysis:

• Tracker Video Analysis – free video analysis software (this is what I use).


• Vernier Logger Pro – not free, but very easy to use data analysis and collection program that also does video analysis.


• Direct Measurement Videos – these are a collection of awesome videos that include data right in the video. Using these, students can analyze motion without any other software.

Ok, but back to unrealistic trajectories. There are countless videos online that show someone or something getting launched through the air. Maybe it is a person jumping off a building or a car driving over a ramp. But in all of these cases, there is most likely just one force acting on the object – the gravitational force.

Objects with only a gravitational force (and near the surface of the Earth) will have a constant vertical acceleration. We call these motions “projectile motion” and they would have the following two equations for the horizontal and vertical directions.

So, if I have a video showing projectile motion and it doesn’t have a constant vertical acceleration and a constant horizontal velocity, then it is probably fake. Oh sure, there could be cases with significant air resistance but for most cases this is an insignificant force.

How about an example? Check out this video from the internet.

Yes. It shows a buggy driving on the moon. THE MOON! Could that be real? Seems unlikely, right? Let’s find out. Like I said before, we need an object that only has the gravitational force on it. In this case, dust from the tires fits the description. Sure, dust on Earth doesn’t fall like projectile motion but on the moon there isn’t any air to keep the dust up.

Using Tracker Video Analysis, I can get the following plots for the vertical and horizontal motions of some of the dust.

The left graph shows that the horizontal velocity of the dust is fairly constant. The right graph shows the vertical motion of the dust has a constant acceleration of 2.14 m/s². So, I think this video is real. Humans actually drove a buggy on the moon. Yes, the accepted value of the acceleration on the surface of the moon is 1.6 m/s² but this value is close and constant. There are some issues with the scale of the video, but it’s still real.

Here is a longer version of this analysis. Also, at the time of my first analysis I wasn’t aware that someone else had done the same thing but I can’t seem to find it again.

You need some more examples here you go.

• Analysis of a Burger King Breakfast Commercial.


• Jumping off a building with bubble wrap.


• Elephants can’t jump.

Of course, there are many many other examples.

Video Analysis Method 2: Impossible Physics

What if I don’t really worry about a frame-by-frame analysis of motion in a video? Instead suppose that I just look at the video to see if it’s even possible based on known laws of physics? Here is an example. Take a look at this video that shows guy at a lab being hit by styrofoam shot from a pneumatic cannon.

From the video, I can get a recoil speed of the lab at about 10 m/s (ok, I lied about not looking at frame-by-frame stuff to get data). If I estimate the mass of the person and the foam projectiles, I can treat this as a standard collision (and conservation of momentum) problem.

If the foam has a mass of 5 kg, it would have to come out of the launcher with a speed of 130 m/s (290 mph) in order to get the kind of recoil shown in the video. What’s even worse is that the guy that was hit would have an acceleration of at least 14.6 g’s. So, if this is indeed a prank it’s one that wasn’t so funny. But it’s fake.

Here is one more example. This isn’t even a video, but just an image of two dudes doing a combo wall climb (you know, where you push on two walls to climb up). But in this case, the two humans are nearly horizontal. This makes the move impossible. There must be some slight tilt such that there a vertical component of a force to balance the downward gravitational force.

Video Analysis Method 3: Detecting Fake Shake

This is my favorite video analysis method. Here is the basic idea:

• Someone wants to make a cool fake video.


• How do record a fake video? One method is to use a camera on a tripod. This makes the background stationary so that some fake video effects can easily be added.


• Real fake videos (yes, I said that) aren’t recorded using a tripod. No, they are recorded on the spur of the moment when something super awesome happens. Tripods imply planning. This can be fixed by adding some fake shake to the video.


• Upload to Youtube.


• Profit.

But wait! Can we detect fake shake? Sometimes. Here is an example video.

Now I can just track the motion of the background. Here’s what that looks like.

Just for comparison, here is the background tracking for a video I made holding a camera as steady as I could.

Notice the difference? The fake video has regular looking “shakes” where as the real video looks more random – actually a lot like a random walk.

Ok, just because a video added a fake shake doesn’t mean the original video is fake. Also, what if the camera was being held by a robot or android? Maybe these artificial intelligences would shake a camera unlike a human – right?

Still, there is a lot of work to be done in the analysis of fake shaking. Is there some metric (like average jump size) that could be used to quantitatively compare real and fake videos? Do different humans have different types of real shake? What does the auto-stabilization feature (either in cameras or from youtube) do to the shake pattern?

Perhaps there will be a whole new field of science – FakeShakeology. There are plenty of videos online to look at and to gather evidence from.

Homework

Of course I am going to assign you homework. Actually, these are the same homework questions I assigned at the end of the Forman Lecture at Vanderbilt University. So, if you already answered these you don’t have to do it twice.

• Find some videos and try a video analysis with each one of these three methods. I suggest you go to this site called “youtube” – there’s tons of video there.


• Explore the natural video shake frequencies for different humans (or find some other metric to measure).


• Make your own fake video. Be sure to make it in such a way that these three methods can not easily detect it as fake. Share it with your friends for fun and profit.


• Build your own fake shake algorithm that takes a tripod video and turns it into a hand-held looking video.