motion-sampler_skill

Overview

This skill extracts frames at regular intervals from dashcam videos to produce compact visual summaries of vehicle movement and location changes. It reduces archive storage by roughly 90% while preserving spatial and temporal context for quick review and trajectory analysis. Outputs include timestamped JPEG frames and a comprehensive INDEX.csv catalog.

How this skill works

The skill scans selected camera folders, skips short or corrupted files, and samples frames starting at 1.0s, then every N seconds until duration-1.0s, always including the end frame. It writes JPEG files with a structured filename and records metadata (original video, timestamp_ms, frame number, size) into INDEX.csv. Processing is parallelized with configurable workers and JPEG quality; future versions add NVDEC GPU decoding.

When to use it

• When you need a visual timeline or motion trajectory from hours of dashcam footage.
• To cut archive storage by 80–95% while keeping representative visual records.
• For quick review of trips to find events or locations without watching full videos.
• When preparing datasets for downstream analysis (map-matching, object detection).
• To create compact trip thumbnails for logs, reports, or cloud uploads.

Best practices

• Use a 10–15s interval for balanced storage vs. temporal resolution; 5s for dense inspection.
• Set jpeg_quality to 85–92 to trade-off file size and visual fidelity.
• Increase max_workers on multi-core systems but reduce on low-RAM machines to avoid OOM.
• Apply date_filter and min_duration to restrict processing and speed up runs.
• Keep output_dir on a drive with ample space and monitor disk-full warnings.

Example use cases

• Extract frames every 10s from all Movie_F videos to create a driving timeline for a week.
• Sample Movie_R videos at 5s intervals for detailed rear-camera review on a specific date.
• Generate sparse samples (30s) from parking cameras to archive months of footage with minimal storage.
• Produce an INDEX.csv for automated ingestion into a mapping or incident-detection pipeline.
• Create visual summaries for fleet reports showing daily routes without storing raw video.