Skip to content

Quality, event cameras, visualisation, benchmark

Frame quality, sync and event cameras

FrameQuality measures the sharpness, the exposure, the noise and two flags (a black frame, a frame repeated from the previous one). Sync estimates the clock offset between two streams by cross-correlating their features (2.9 ms error on 20 seeds against a uniform offset in [-50, +50] ms). EventFrame accumulates an asynchronous event stream into a frame with the polarity sign.

Sharp, blurred and black frames with their sharpness, noise and flags.

Visualisation

FalseColor brings any band (thermal, depth, NIR, UV, SAR) into a visible image with a colormap; Overlay draws the tracks, the mask, the flow and the score on a frame; VideoWriter writes the annotated video and one summary image with the signals in time.

A thermal band in false colour and the same frame with a box drawn on it.

Benchmark

benchmarks/video_quality.py measures every module above at 640x360 on three synthetic streams (a smooth motion, a sudden cut, a multi-channel thermal sequence). The JSON next to the script carries the numbers and the machine. The 1280x720 and 1920x1080 runs exceed the free Colab session, and are reported as not measured.

Median milliseconds per frame at 640 by 360 for every module of the video package.

API