Peak action detection
Also known as: apex detection, action peak finding
Software locating the moment of greatest activity inside a clip, so a long recording can be trimmed around it automatically.
Updated
What it means
Detection answers a narrow question: where in this file is the most happening. The output is a timestamp rather than a judgement about quality. One file, one position, nothing more than that.
Motion is the usual signal. Motion vector analysis measures how much the picture changes between frames, and a sharp rise marks something worth looking at more closely. Smooth motion is as informative as sudden motion.
Audio contributes as much as picture. An Audio event detection spike from an impact or a shout frequently locates a moment the image alone would miss, since sound fills a whole frame.
How it works in practice
It is most useful on long continuous files. A forty-minute mounted recording has no clip boundaries, so detection is what turns it into candidates and Segment scoring is what ranks them afterwards. Nothing in the file marks where the runs were.
Thresholds decide the yield. Threshold tuning trades missed moments against false ones, and the right setting depends on whether the session was dense or quiet, which no default setting can know in advance.
Combining signals works better than any single one. Motion, audio and Shot change detection agreeing is a stronger indication than any of them alone, and disagreement is informative too.
What to watch out for
Motion is not interest. A camera shaking on a rough descent registers as enormous activity, and a Blur detection check is what stops that becoming a selection.
The peak of motion is not always the readable frame. A crash peaks at impact, where the clip that plays well starts before it, so detection locates a region rather than a cut point.
Quiet moments are invisible to it. A long carve, a held line or a calm aerial have no spike at all, which is why Shot scoring has to judge more than movement.
How RawClip handles it
Detecting where the action sits is the first step of the analysis, and it runs on motion, audio and scene changes together rather than on one signal. Scoring then decides which of those moments are worth keeping, which is a separate judgement from finding them.