A Flight-Parameter-Based MLP Model for Motion Vector Prediction of Video Taken Onboard a Drone

Şimşek A., DÜNDAR G.

2026 IEEE International Conference on Consumer Electronics, ICCE 2026, Dubai, Birleşik Arap Emirlikleri, 3 - 05 Şubat 2026, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/icce67443.2026.11449920
  • Basıldığı Şehir: Dubai
  • Basıldığı Ülke: Birleşik Arap Emirlikleri
  • Anahtar Kelimeler: drone, flight parameter modeling, H.266, motion compensation, motion estimation, motion vector prediction, VVC
  • Boğaziçi Üniversitesi Adresli: Evet

Özet

While the use of inter-prediction and bidirectionally coded (B) frames improve compression efficiency during video coding, due to the use of look-ahead buffers and out-of-sequence coding, it may not suit some delay sensitive practical applications. During inter-coding similarity search based general-purpose algorithms are utilized for motion estimation which are computationally complex. If future motion can be predicted from external metadata, encoding can be optimized for lower complexity. In this study, we propose a multi-layer perceptron (MLP) based motion model for predicting the motion vectors of video taken onboard a drone using the flight parameters of it. The results show that flight-parameters-aware motion vector prediction for motion estimation provides us with reduced computational complexity during drone video coding.