Video Live Streaming Quality Issues, Evaluation Methods, and System Design

Live video streaming often suffers from quality degradations such as blurriness, block artifacts, color bleeding, delay, and stuttering, which differ from the ideal scenario where viewer and broadcaster see identical video.

Typical quality problems are grouped into codec‑related (e.g., blur, motion artifacts), network‑related (e.g., latency, buffering), and cross‑related effects (illustrated by the included image showing blockiness and false edges).

Root causes include differing source cameras, varying compression standards, and different encoder parameters (bitrate, frame rate, resolution), as well as network conditions and playback devices.

Quality evaluation methods are divided into subjective and objective approaches. Subjective evaluation uses human viewers to obtain a Mean Opinion Score (MOS) following ITU‑R BT.500 or ITU‑T P.910 standards, but it is costly, time‑consuming, and subject to fatigue and bias.

Objective evaluation employs mathematical models that simulate subjective results without human involvement. It is categorized into:

Full Reference (FR): compares each pixel of the decoded video to the original, using metrics such as PSNR and SSIM, though their correlation with MOS varies.

No Reference (NR): estimates quality from the decoded video alone, which can be inaccurate when no reference is available.

Reduced Reference (RR): extracts and compares selected features from both original and decoded streams, useful when bandwidth limits full reference transmission.

Both subjective and objective assessments are often combined; objective metrics guide algorithm development, while subjective tests validate perceived quality.

Role of video quality assessment includes guiding codec improvements, informing hardware and software selection, and supporting higher‑level applications like streaming platforms, live broadcast, and video communication.

Design of a live‑streaming quality assessment system involves using identical video inputs, applying pre‑processing, encoding, transmission, decoding, and post‑processing, then recording side‑by‑side outputs for comparison. The setup requires appropriate cameras (mobile or HDMI), processing and transmission tools, network impairment simulators (e.g., Linux Traffic Control), high‑resolution displays, and multi‑stream recording hardware such as Blackmagic devices.

Additional auxiliary equipment like a magnetic levitation globe (for consistent motion) and a millisecond‑accurate electronic clock (for latency measurement) can enhance test repeatability.

The testing environment should mimic real‑world consumer conditions rather than strict laboratory settings, using realistic lighting, subjects, and content.

Personnel selection depends on the test goal: experts for codec development, non‑experts for consumer‑level product comparison.

The evaluation workflow for an entertainment live‑stream scenario involves four smartphones (two broadcasters, two viewers), side‑by‑side recording on a 4K monitor, network condition manipulation, and both objective data collection (latency, frame rate) and subjective scoring using a five‑point scale.

Overall, the described system provides a repeatable, realistic, and comprehensive framework for assessing video live‑streaming quality across diverse conditions.