ForensicSearch 02Technical ArticleGaitAI Research2 min read
How GaitAI Compares Two Walking Videos
Two street recordings show people walking at a similar pace. The comparison starts with the footfalls each camera captured.
When two cameras capture different portions of a walk, the first step is to find something visible in both recordings. In this ForensicSearch demonstration, GaitAI starts with the moment a shoe touches the pavement.
Start with what each camera captured
Both cameras show a side view, but the walkers move in opposite directions across the screen. A reviewer can use visible foot contacts as common reference points, even when the angle, background and length of the recordings differ.
Find a complete walking cycle
Follow the white shoe from one contact with the pavement to the next contact of that same foot. That interval is one gait cycle. Recording A shows two complete cycles and Recording B shows five. The counts tell the reviewer how much of each walk is available for comparison.
Compare the movement from two views
Pose estimation follows the positions of joints such as the hip, knee and ankle. Silhouette extraction traces the outline of the body through the same frames. Together they let the reviewer inspect individual joint movements and see how the whole outline changes.
Cadence adds a measure of pace. Recording A shows 99.2 steps per minute and Recording B shows 96.4. The similar cadence gives the reviewer one point of comparison alongside the visible steps and body outlines.
Check whether the legs can be tracked reliably
Recording B has inconsistent left and right leg labels. A measure of step-time asymmetry needs a reliable assignment of each footfall to a leg. When that assignment can change from frame to frame, the comparison leaves the asymmetry unreported.
Return to the source footage
The cycle counts and cadence give a reviewer specific observations to check against the original clips. A reviewer can replay the relevant footfalls, inspect where the tracking becomes uncertain and interpret the movement alongside the other evidence in the case. Explore ForensicSearch to see the workflow for finding and reviewing candidate footage.
The technical walkthrough shows how the two walks are aligned and how a series of outlines becomes a gait energy image.
Request a Demo →