An 8×8-block based motion estimation using Kalman filter

Kalman filtering based motion estimation for video coding with adaptive block partitioning

“…To further improve the performance of the Kalman filter, a zigzag scanning of the blocks is proposed and state parameters of the Kalman filter are adjusted adaptively during the each KF iteration. Simulation results show that the method developed in this work yields more accurate motion vectors (MV) and better motion compensated images in terms of PSNR than the methods proposed in [4] and [5], respectively. A benefit from the KF is the fraction pixel accuracy of motion vectors with no additional bit rate for MVs.…”

“…In [4], the non-adaptive and adaptive Kalman filtering (KF) methods were proposed to enhance the motion estimates based on 1-D and 2-D autoregressive models. An 8x8-block based motion estimation algorithm was proposed in [5] which uses the KF technique to improve the motion estimates resulting from three-step search algorithm. The Kalman filter was employed in [6] to enhance the motion estimation performance for very low bit rate video coding.…”

“…To improve the motion estimation performance, macro-blocks were split into 8x8 blocks in [5] for the Kalman filtering. In our work, we propose a novel method so that the macro-blocks are split into 8x8 blocks or 4x4 sub-blocks adaptively for the KF as shown in Fig.…”

Establishing target track history by digital watermarking

“…In its simplest form tracking is defined as the process whereby a target is first detected and then tracked from frame to frame. Existing techniques include block matching, 1 Statistical inference, 2 Kalman filtering, 3 Hidden Markov Models 4 and more recently Particle Filters. [5][6][7] None of these algorithms by themselves address track history question and spatio-temproal relationship among tracked targets.…”

A real-time motion estimation FPGA architecture

“…Among the most demanding tasks in video encoding is the Motion Estimation (ME) process, because it requires more memory space and it is computationally more intensive when compared to the other processes performed in the encoder. Targeting the improvement of the Motion Estimation performance a number of algorithms have been presented in the literature [1][2][3][4][5]. The effort in producing new algorithms has targeted the improvement of the computational complexity while keeping the resulting QoS as close as possible to that of the Full Search Algorithm [6].…”