Deinterlacing is the process of converting old interlaced-scan video (where each frame was split into two half-frames called fields) back into modern progressive-scan video (where each frame is one complete image). Without deinterlacing, interlaced material played on a modern flat-screen TV or computer shows the famous "combing" artefact — visible horizontal "teeth" along anything that moves.
The work is harder than it sounds because the two fields of an interlaced frame were captured a fraction of a second apart. To reconstruct a single clean progressive frame, the deinterlacer has to either pick one field and throw the other away (cheap, halves vertical resolution), blend the two fields (cheap, gets blurry on movement) or properly understand the motion between fields and interpolate (expensive, much higher quality). Good deinterlacers like Yadif, BWDIF or motion-compensated commercial tools track moving objects between fields to reconstruct what each one "should have looked like" at the right moment.
For a modern product, deinterlacing only matters at the ingest end of the pipeline. Anything captured in the last 15 years on a modern camera is already progressive; you only encounter interlaced material when ingesting archival broadcast tapes, old DVDs, legacy IPTV feeds or some satellite signals. The standard answer is: detect interlacing automatically (FFmpeg's idet filter), apply a good motion-compensated deinterlacer once at ingest, then never touch the material again. Deinterlacing on-the-fly in the player is a bad idea — it shows the work poorly and most modern players don't even attempt it well.
