Uppsats

Exploring Intra-Frame Compression for Interactive After-Action Video Review : A Comparative Evaluation of H.264, MJPEG, and JPEG XS

Kandidat-uppsats

Linköpings universitet/Institutionen för datavetenskap

Publicerad: 2026

Språk: Engelska

Sammanfattning

Video material in simulation-based training systems is often recorded and later reviewed during After-Action Review (AAR). To make this review process interactive, the encoded video should support responsive playback, random access, and faster-than-real-time navigation while still maintaining acceptable quality and manageable bandwidth and storage requirements. This thesis compares H.264, MJPEG, and JPEG XS as intra-frame compression alternatives for interactive AAR. The evaluation combines benchmarking with frame-markerprototype implementations for identifying safe access points in compressed video streams, meaning positions where decoding can begin without depending on earlier frames. A stand-alone Java prototype was developed to generate frame marker lists for MJPEG and JPEGXS, while FFmpeg-based benchmarks were used to compare bitrate, storage usage, encoding and decoding performance, CPU time, and visual quality measured with the objective metrics SSIM and PSNR. The results show that intra-frame compression simplifies frame-level random access compared with inter-frame coding, but increases bandwidth and storage requirements. Among the tested codecs, H.264 intra provided the strongest compression efficiency, MJPEG achieved the fastest encoding performance and allowed simple marker-based frame detection, while JPEG XS offered predictable bitrate behavior and strong decoding performance at a higher bitrate cost. Frame markers were implemented for both MJPEG and JPEG XS. MJPEG frame boundaries could be identified through direct marker search, whereas JPEG XS required additional codestream validation to avoid false detections. The results indicate that the suitability of a codec for interactive AAR depends on the trade-off between compression efficiency, processing cost, predictable transmission, and practical support for random access.

Information

Författare
Tran, Robert
Lärosäte / institution
Linköpings universitet/Institutionen för datavetenskap
Publiceringsdatum
2026
Uppsatstyp
Kandidat-uppsats
Språk
Engelska