Getting Started With Tracker For Physics Weekly
Tracker is a free open-source video analysis tool originally built by Bruce Sherwood and Ruth Chabay at Texas A&M. It lets you track objects frame by frame in video files, generate position-time and velocity-time graphs, and fit mathematical models to your data. If you are following Physics Weekly for problem sets or lab recommendations, this is the software most people end up using alongside it. The install is straightforward. Download it from [physlets.org/tracker](https://physlets.org/tracker) and run the installer for your operating system. No account required. The default settings are fine for basic work, but there is one thing most people miss: the coordinate system orientation. By default Tracker sets positive y upward and positive x to the right, which matches standard textbook convention but will produce inverted results if you drag-and-drop a new point after flipping the video. I ran into this exact issue when a student submitted a projectile motion lab where every velocity component had the wrong sign. The fix was to click View and check the coordinate system box before marking any points. Takes two seconds and saves an hour of debugging.
Tracker For Physics Weekly Workflow
Here is how I actually use it week to week when Physics Weekly drops a new analysis assignment. Import the video first. Drag it into the window or use File > Open. Most assignments provide a direct download link. Wait for the full frame count to load. The playback range defaults to the entire video, which is usually correct, but verify by scrolling to the end of the timeline. You should see the last frame render properly. If it stops early, the codec may be unsupported. MP4 with H.264 encoding works reliably. MOV files sometimes cause issues on Linux builds, so convert those first with ffmpeg. Calibrate your scale next. Click the Calibrate button, then use the stick tool to mark two known points in the frame. Type the actual distance between them. This sets your pixel-to-meter ratio. Without this step every measurement is meaningless. I have seen people skip this and try to work in pixels, then get confused when their acceleration values look nothing like 9.8 m/s^2. The calibrated scale shows up in the bottom left corner of the video window. Verify it reads correctly before moving on.
Set up your coordinate system. This is where I made that earlier mistake. Use the axis tool to draw x and y directions that match your physical setup. For an inclined plane experiment, the x axis should run parallel to the ramp surface, not horizontal to the screen. Flipping the axes later is possible but tedious because all your mass points need manual re-indexing. Get it right the first time. Mark your object. Click Auto-Track or mark each frame manually. Auto-Track works well for high-contrast objects against a uniform background. If your object has variable lighting or partially leaves the frame, Auto-Track will lose it and insert garbage data points. When that happens, switch to manual tracking and work in shorter segments. A 1080p video at 30 fps of a 5-second experiment gives you 150 data points. Manual tracking takes roughly 15 to 20 minutes depending on your mouse control. Not terrible, but you do not want to do that every week. Generate graphs from the data table. Once your mass points are marked, go to the Data Table panel and click the graph buttons. Position vs. time and velocity vs. time come up automatically. You can overlay multiple tracks on the same graph. This is useful when comparing a theoretical prediction to your measured data.
Get the Full Details
Fit a model. Click the curve-fit button and choose the function type. Linear for constant velocity. Quadratic for constant acceleration. Exponential for damping cases. The fit coefficients appear in the output window with their uncertainties. Pay attention to the reduced chi-squared value. A value significantly above 1.0 means your model does not fit the data well, or your error bars are too small. A value near zero usually means you overestimated your measurement uncertainty. Export your results. File > Export lets you save data as CSV, graphs as PNG, or the full project file. When submitting for Physics Weekly assignments, export both the data table and the graph. Some graders want to see the raw data to verify your tracking quality.
Where Tracker Falls Apart
It is not a perfect tool. Here are the places it genuinely struggles. 3D motion is impossible to track accurately. Tracker operates in two dimensions. If your experiment involves anything out of the plane of the camera sensor, your measurements will be compressed and distorted. No amount of calibration fixes this. Use photogrammetry software instead if you need true 3D tracking. It costs money and has a steep learning curve, but it works where Tracker cannot. High-speed cameras produce data density problems. At 1000 fps, a 10-second clip gives you 10,000 frames. Tracker can handle it, but the interface becomes sluggish. I usually downsample to 120 fps for analysis unless I specifically need the temporal resolution. The tradeoff is negligible for most introductory physics experiments and the workflow becomes significantly faster.
Automatic tracking fails on complex backgrounds. I worked through a lab once where students filmed a pendulum swinging in front of a classroom window with moving clouds. Auto-Track locked onto the clouds instead of the pendulum bob. Had to switch entirely to manual marking, which took about 40 minutes for a 3-minute video. The workaround here is to place a contrasting background behind your object before filming. A simple sheet of colored construction paper costs nothing and eliminates the problem entirely. Camera distortion is ignored. Wide-angle lenses introduce barrel distortion that warps positions near the edges of the frame. Tracker does not include a lens distortion correction feature. If you are doing precision work, keep your object near the center of the frame and use a telephoto setting if possible. Linux users may encounter driver issues with newer GPUs. The video rendering pipeline sometimes conflicts with proprietary NVIDIA drivers. The official Linux build is tested against Nouveau and Mesa drivers. If you are on NVIDIA proprietary, run `MESA_GL_VERSION_OVERRIDE=3.3` before launching Tracker as a workaround. It forces compatibility mode and the video playback stabilizes.
Common Pitfalls to Avoid
Not zeroing your initial conditions. If your object starts at a non-zero position, the integration constants in your model fit will absorb it, but you need to account for it when comparing to theory. Write down your initial position and velocity before you start tracking. They matter more than people expect. Using too few data points. Five or six marks across a 5-second video gives you a terrible fit. Aim for at least one mark per tenth of a second. More is better. The extra effort pays off in tighter confidence intervals on your fitted parameters. Forgetting to set the frame rate. Tracker defaults to 30 fps if it cannot read the frame rate from the video metadata. If your video was encoded at 24 fps or 60 fps, your time axis will be wrong and every derived quantity will be scaled incorrectly. Check View > Video Properties and confirm the frame rate matches your recording device. This single oversight is the most common source of error I see in student labs.
Ignoring uncertainty propagation. Tracker reports fit uncertainties but does not propagate your tracking error into those numbers. If your pixel resolution limits position accuracy to about 2 pixels, your velocity uncertainty is roughly 2 pixels divided by the time between frames. Factor this into your final error bars manually. Your grader will notice if you do not. If you want an alternative, Vernier Video Analysis is commercially supported and has a cleaner interface, but it costs money and requires a subscription for cloud features. For anyone doing casual or educational analysis, Tracker covers the vast majority of use cases at zero cost. The friction is worth it.
