Automated Video Quality Detection and Multithreaded Optimization for Live Stream Transcoding

Background : In live‑stream processing, after media service transcoding, the resulting video stream must be validated. Manual verification is error‑prone, so an automated quality‑check pipeline was introduced.

Core Process : The overall task includes (1) obtaining the live source, transcoding and pushing to CDN, (2) capturing screenshots, and (3) performing quality checks for black‑screen or artifact issues.

Quality‑Detection Flow : Frames are extracted from the transcoded stream and sent to an internal testing platform. Images are base64‑encoded (URL‑safe) and submitted via API for analysis.

Image Capture Implementation : Initially, ffmpeg was used for screenshots, but its process conflicted with the live‑push ffmpeg instance, causing interruptions. The solution switched to OpenCV for independent image capture, ensuring the capture process does not block the main workflow.

Result Storage : Captured images are uploaded to the company’s S3 bucket so that remote test reports can reference them directly, avoiding local‑only storage limitations.

Thread‑Based Optimization : A custom threading.Thread subclass provides a get_result() method to retrieve detection outcomes. The main flow waits for the detection thread via thread.join() . To accelerate detection, the workload is split across multiple threads, each handling a subset of images.

Producer‑Consumer Model : After determining the number of capture and detection threads (based on task duration and stream count), a global queue holds captured frames. Detection threads consume from the queue, enabling a “capture‑while‑detect” approach that eliminates idle time.

Performance Evaluation : For a typical 30‑second task at 15 fps (450 frames, 30 captured images), single‑thread detection takes ~174 s, while multithreaded “capture‑then‑detect” reduces it to ~38 s, and the “capture‑and‑detect” pipeline further cuts total time to ~28 s, a 42 % improvement.

Conclusion : By integrating automated quality detection, OpenCV‑based capture, S3 storage, and a multithreaded producer‑consumer architecture, the system achieves reliable, fast video quality validation without impacting the primary transcoding workflow, providing a reference for future similar integrations.