Press "Enter" to skip to content

2D Video Analysis

Jonas Ebbecke 0

Last updated on 30. April 2020

Biomechanical investigations – many people think of laboratories equipped with marker-based camera systems from Qualisys or Vicon that can record movements in three dimensions. However, 2D video analysis is often completely sufficient. This saves time and, above all, money. Contrary to what some people may claim, considerably more parameters can be derived from normal video material than just joint angles or segment speeds.

Your job as a biomechanist is to quantify and analyse movements. In the first place it does not matter if you are a coach or a scientist. A simple but very effective way to do this is the 2D video analysis. You can use it for any kind of planar movement. For example: running, long jump, cycling, javelin throwing, gymnastic elements, high diving and much more. In consequence, movements that take place in several planes (such as dancing or high jump) are unsuitable and must be analysed in three dimensions.

For a scientific 2D video analysis you only need five things: camera, storage medium, calibration frame, markers and a PC with an analysis software.

When choosing a video camera for biomechanical analysis you should consider the following things:

  • 1.   Sufficient image quality
  • 2.   Sufficient frame rate 
  • 3.   Manual focus
  • 4.   Manual shutter speeds
  • 5.   Manual aperture
  • 6.   Longest possible focal length
  • 7.   Saving files in .raw format without video compression
  •  

Due to the technical progress of the last few years, almost everyone today has a camera in their smartphone which has more than an adequate image quality. However, if you take a look at the frame rate, mobile phone cameras that record at 30-60Hz as standard sometimes reach their limits. The faster the movement, the higher the recording frequency should be. Nyquist has established a rule of thumb:

f(record) > 2 * f(signal)

Therefore, you should think carefully about what part of the movement you want to analyze and purchase a camera with sufficient frame rate. However, you do not need a high-speed device with a recording frequency of 2000Hz for every movement. These are very expensive, the recordings take up a lot of memory space and the post-processing takes much more time.

For most biomechanical analyses, fast shutter speeds are required, otherwise the image will get blurry for fast movements. This leads to complications in post-processing. But again, the optimal settings for the individual examination must be found. Because: the shorter the exposure of the sensor, the darker the image becomes. Consequently, a too small shutter speed could also lead to complications in post-processing. If this is the case, you can either adjust the aperture or use artificial light sources.

Last but not least you should choose the longest possible focal length. This is the distance between lens. A long focal length prevents perspective shifts at the edges of the image, as you certainly know from wide-angle cameras like the GoPro. Such shifts have an enormous influence on the analysis of the image and the derived parameters. However, the focal length is limited by the fact that it involves a zoom factor. This means that the longer the focal length, the greater the distance between the camera and the subject has to be, so that the entire movement can be captured. Therefore, the size of the focal length is often limited by the room where the analysis is performed.

Position the camera correctly.

Usually, the camera remains in one position and the subject moves through the image. This makes it easy to determine the subject's movement parameters relative to an external reference system after calibration (see below). However, there are also special types of video analysis where the camera moves. This can be used for longer sequences of movements, such as a sprint, in order to analyze the whole event. Depending on the movement of the camera relative to the outer reference system (rotational or translational), mathematical corrections must be made for this movement if you want to obtain accurate two-dimensional coordinates.

You should also align the optical axis of the camera perpendicular to the plane of motion. All movements that occur in a plane parallel to the image sensor of the camera are not subject to any perspective error during the digitization phase. However, no human movement is really planar. Therefore, you should aware in advance which aspect of the activity is of primary interest and in which plane this will happen. The camera can then be positioned accordingly. A reference shot of a right-angled triangle can help you to find the correct positioning of the camera. Unfortunately, even a few degrees in the camera orientation can have a big impact on the accuracy of the analysis.

You should therefore position the camera as far away from the athlete as possible. This will reduce the perspective error caused by movements outside the performance plane. With a telephoto lens, which has a long focal length, you can still take your desired picture size. However, you should only use an optical zoom. If you choose a digital zoom (like your smartphone camera does), the image quality will be reduced. The ideal image size is determined by the motion to be analyzed. Make sure that the whole movement is covered, but don't waste image space!

Select the correct camera settings.

The frame rate used is determined by the frequency of the motion to be analyzed. Nyquist has established a rule of thumb for this: The sampling frequency (frame rate) must be at least twice as high as the highest frequency present in the activity itself. In reality, however, the frame rate should be significantly higher than this (we suggest 8-10 times higher). A sufficiently high frame rate ensures that key events of an activity (e.g. foot strike during running or ball impact during a tennis serve) are recorded. Increasing the frame rate also serves to improve the accuracy of time measurements. This is especially important when the phases are of short duration. Here are a few example frequencies:

  • 25-50 Hz - Walking, swimming, stair climbing.
  • 50-100 Hz - Running, shot put, high jump.
  • 100-200 Hz - sprinting, javelin, soccer kick.
  • 200-500 Hz - tennis serve, golf swing, parry in fencing.
  •  

In most activities, the furthest segments of the body, i.e. the hands and feet, move the fastest. The shutter speed should be set, that even these segments are not recorded blurred. Therefore, the choice of shutter speed also depends on the type of movement. For slow activities, such as walking, shutter speeds of 1/150-1/250 seconds should be sufficient. For slightly faster movements, such as running or a swim start, shutter speeds of 1/350-1/750 seconds are more appropriate. Very fast activities such as a baseball hit should be recorded at shutter speeds of 1/1000 second or faster. You should note, however, that increasing the shutter speed will always be accompanied by poorer lighting conditions. If they are no longer sufficient, you must provide additional artificial light.

Furthermore the camera should be set to manual focus mode in most cases. For a well focused image, zoom in completely on an object in the plane of motion, focus manually and zoom out to the desired field of view.

Camera calibration is about transforming the image coordinates into spatial coordinates (i.e. coordinates of the real world) after digitization. For this you need a scaled object whose height and width you know exactly. You have to record this object in the plane of motion with your camera. To minimize the scaling error, you should choose the dimensions of the scaled object so that they take up a large part of the field of view.

After capturing the object, you can use the so-called fractional linear transformation (FLT) to determine the conversion factor:

S = real length / digitized length [m/px]

x = S * u

y = S * v 

[x,y: real-world coordinates; u,v: image cordinates]

In some setups it is not always possible to align the optical axis of the camera perpendicular to the plane of motion, e.g. for analyses during competitions. This can be corrected by means of the so-called two-dimensional direct linear transformation (2D-DLT). This has been shown to produce significantly more accurate reconstruction of two-dimensional coordinate data (Brewin and Kerwin, 2003).

It is important that the athlete wears clothing that allows you to identify the relevant anatomical landmarks during your analysis. In addition, small markers on the subject's skin can help to locate these body parts during digitization. The positioning of the markers should be well thought out!

Due to the movement of the soft tissue relative to the skeleton, there may be shifts between the skin marker and the anatomical landmark during activity.

The number of recorded attempts depends on the purpose of the analysis and the level of the participants. Because the movement patterns of trained athletes are likely to be much more consistent than those of inexperienced athletes, they may need to perform fewer attempts to demonstrate typical performance. The more test subjects and individual attempts you include, the more important it becomes to keep the overview. Sometimes it can be helpful if you record a piece of paper, a board or something similar with the corresponding test person and test number written on it directly before the respective recording.

And finally, the most important thing to save time and nerves: Make sure that nobody else touches your camera or changes the settings! Otherwise you may start all over again 😉

In order to analyze your recordings you must first digitize them. That means you mark the anatomical landmarks in each frame and save their image coordinates. With the help of the FLT described above you can now convert the image coordinates into two-dimensional space coordinates.

Using the selected anatomical landmarks you can now create a model, i.e. a strong simplification, of the body. In biomechanical models the human body is usually divided into rigid body segments, which are connected by joints. With the help of the frame rate, it is now possible to calculate, where each body segment is located at a specified point in time and at what angle it is positioned relative to the other segments. In addition to the anatomical landmarks you should also know the body weight of your test person. Applying a body model (for example Winter, 1991) assigns an empirically determined proportion of the total body mass, a center of mass and inertial properties to each segment. This allows a whole range of parameters to be calculated:

Kinematics:

  • 1.    Position, velocity and acceleration of individual segments or the centre of gravity
  • 2.   Joint angle, segment angle, angular velocity and acceleration
  • 3.   Contact times
  •  

Kinetics (inverse dynamics):

  • 1.   Momentum, kinetic and potential energy
  • 2.   joint moments, internal forces
  •  

That all sounds like a lot of calculating - and it would be! But fortunately we live in the technological age and do not have to do these calculations one by one. There are some free programs that have been programmed for two-dimensional motion analysis, like SkillSpector or Kinovea.

Using a long jump video as an example, we show you how to perform a 2D video analysis.

References

Payton, C. (Ed.), 2017. Biomechanical evaluation of movement in sport and exercise. The British Association of Sport and Exercise Sciences guide. Routledge, London.

Brewin, M.A., Kerwin, D.G., 2003. Accuracy of Scaling and DLT Reconstruction Techniques for Planar Motion Analyses. Journal of Applied Biomechanics 19, 79–88.

Winter D.A, 2005. Biomechanics and Motor Control of Human Movement, 3rd edition. John Wiley & Sons.

Leave a Reply

Your email address will not be published. Required fields are marked *