Unlimited seeking during videotranscoding

Abstract: High definition, long duration videos can contain complex scenes full of rich detail. As the details get richer and the duration grows, the cost of transcoding invariably increases. For Swedish law enforcement, a greatly detailed video can prove vital in aiding the investigation of a crime, by clearly identifying objects, or the face of a perpetrator or victim. Modern day surveillance cameras are capable of capturing videos with high frame rate and great resolution, but the interesting scenes usually make up only a small fraction of the recorded material. Finding those scenes quickly means being able to search through the entire video unhindered. To aid in this process, a novel web application was developed during this thesis that allows users to browse videos on a server, load them into a video player and seek across the entire video while it is being transcoded, regardless of the original video format. Experimental results identify the CPU as the primary bottleneck, and a preferable FFmpeg transcoding preset (called veryfast) from a setof candidate presets. Regarding usability, the application runs mostly without issues, but long buffering times will sometimes plague the user experience. Yet, this thesis provides an important step towards achieving seamless and simultaneous seeking and transcoding.