Application Programs and Systems - Heterogeneous Multicore Processor Technologies for Embedded Systems - page 188

Small Data
Volume Application

Event Detection

S-T MRF Algorithm

Thread
Parallel
Processing

Object MAP Creation

Motion Vector
Extraction
( SAD Operation)

Large Data
Volume

Fig. 6.15 Structure of S-T MRF application

Software Layer

Application

Frame Pointer Pixel Pointer

Event
Detection

Object MAP
Creation

Motion Vector
Extraction

Object MAP Vector

MX Core

CPU#1 CPU#0

MPC

MPA

Hardware Layer

Fig. 6.16 Thread parallel processing scheme

Figure 6.17 illustrates the frame pipelining technique for the thread parallel pro-
cessing shown in Fig. 6.16 . The horizontal axis is the time, and the time unit refers
to the processing time of each frame. Each output of the thread is given to another
thread that is processed one time unit later. Parallel execution of the threads is
achieved with this frame pipelining technique.
In the S-T MRF algorithm, the motion vector extraction is based on the sum of
the absolute difference between the sequential frames. The SAD algorithm is very
useful for evaluating the similarity between two frames, as depicted in Fig. 6.18 .