Video Camera Options for video analysis

You have no idea how long I have been sitting on this one. I made some videos like a billion years ago, and still no post. Why? Oh well, here it is.

I like video analysis of motion. I like looking at stuff on youtube or other video sites. But sometimes, you need to make the video yourself. What should you use? My personal favorite is a Flip Mino HD. It is small and quick. Are there other options? Sure. In this post, I will do the video analysis with the following cameras:

  • Flip Mino HD
  • A Canon PowerShot A470
  • Panasonic DMC-FZ18
  • Canon HV10 HD miniDV camcorder (you know, the kind that uses miniDV tapes)
  • LG Shine phone camera
  • Built in iSight camera in a Macbook Pro

For this test, I set up all the cameras together and had them record the same event. This is me in my garage throwing an oversized tennis ball I stole from my dog. At the time of the test, I couldn't find a silly meter stick, so I used a yard stick.

Flip Mino HD

Here is a shot from Tracker Video:

i-9e334ed953c5ac417a29797a7bcfb6f7-data_tool1.jpg

You can see it gives an acceleration of about 9.8 m/s2. Good for me. There is some interlacing issues - I will show a comparison frame for each camera at the end.

Canon HV10 HD miniDV

I imported and exported this through iMovie, so it is not raw. However, it does give a very good, clear and crisp image. Here is the data from Tracker.

i-023e30a49a207f4ce7adb9dc5c206756-data_tool_11.jpg

Notice that this one gives an acceleration closer to 10.2 m/s^2. I suspect this depends on the scaling of the video (but I didn't do an error estimate calculation).

Built in iSight

Here you can see a difference from the other cameras. The image quality is not very good, it is difficult to see exactly where the ball is in each frame. Also, there seems to be a problem with the frame rate. A couple of times there would be duplicate frames. You can see this in the plot from Tracker.

i-e30d05c3b9844a926a09942322f54a5f-data_tool_2.jpg

Even though the plot doesn't look like a pretty parabola, it still gives a reasonable acceleration.

LG Shine Phone

First, the format of this movie was not compatible with Tracker Video. I used MPEG StreamClip to convert it to a .mov format. Honestly, I could barely see the ball in each frame. I felt like I had the blast shield down and was stretching out with the force to find the location of the ball. Here is what I got.

i-13d5caf7caaef18ab5436baf694a9f05-data_tool_3.jpg

I must be strong in the force.

Panasonic DMC-FZ18

This one made pretty good videos despite being still shot camera.

i-20d27910c00419d0abd0fb4d5c57ad57-data_tool_4.jpg

Canon PowerShot A470

i-785c0aebb08e5dcc59924052f3554d1f-data_tool_5.jpg

Not bad.

Which ones?

Really, all of these worked to some degree - even the stupid cell phone. The question comes down to speed. Warning: I am not a video expert, so what I am saying might be completely made up stuff. Ok, you have been warned. Many video cameras use interlaced video. This means that it takes every odd line of scanning and then every even line. If things are moving slow, there is not much change between the time it takes these two partial pictures. If things are moving fast, there is a difference. However, for normal video usage, viewing fast moving objects looks ok. The problem comes about when you look at one frame. Here is an example from my videos:

i-640112cec712a3df9c0ae88bb1fb58f1-untitled_4.jpg

Here the only ones that clearly show the ball are the Flip and the HV-10. I think the panasonic would have looked better, but for some reason the white ceiling light made everything very bright. These videos show a point where the ball is moving fast. At the highest point of the trajectory, the ball is moving slow and looks much better in all the videos. I am not sure if the Canon HV-10 is progressive scan, but that means that it scans both fields at the same time. Progressive scan = good.

There is a way to fix some of these videos. You can have program "average" the two interlaced frames. This is (unsurprisingly) called "deinterlaced" video. I use MPEG StreamClip to do this. Sometimes it can make a big difference. I tried this with the video from the powershot and it looked about the same.

I think the moral of this post is that a better camera is better. However, you can still find useful things to do with even a cell phone camera. In fact, I have a video from the fair that has been sitting in my "to do list" for quite some time. Sometimes though, you need a good camera to get anything useful. In fact, I had another set of videos that I shot outside. The ball was not very visible in the background though. For that situation, only the HV-10 or the Flip could get useful data.

More like this

I just had a thought to share about why some of the motion paths might be jerky. I had been trying to accomplish a video analysis manually (a meter stick next to a falling ball and reading the frame-by-frame time off of the player - I only just discovered Tracker). In the process I found that my cell phone uses a variable frame rate. Most software would report the average frame rate, but there was one program that gave more detailed info. Also, flipping through frame by frame and counting confirmed that there were less, or more frames in some seconds than in others. But more importantly, the frames and the time codes didn't match up when you got down to sub-second intervals.
Here's what I think is going on: The cell phone snaps frames as fast as it can, and there is some variation in how fast it can do this. It marks certain key frames (let's say the first frame in any given second) and evenly distributes the others. But if the interval between frame #5 and #6 happens to be be rather long, and all the frames get redistributed evenly, it will look like the object was going abnormally fast between those frames. (That happens to be exactly what happened during my test video. That or the ping pong ball was going 40 m/s after a 1 second fall.)
The only thing I found that helps is making sure that if the file needs to undergo a format conversion, make sure the frame rate of the output is the same as the source - several converters insisted on changing it to 25 or 30 fps, or to the average fps. This process invariably messed up the timing. Handbrake (and as a second choice, Freemake) have been working well for me.
But I never found a way to work around having the wrong frame duration other than to take multiple videos and use the one that doesn't have a "hiccup" right at the beginning or end of the motion.

By Paul Bunnell (not verified) on 29 Jun 2011 #permalink