For performing the intensive image data processing
algorithms, general-purpose computing on graphics processing
units (GPGPU) can be an excellent candidate because the data
of feature detection phase is good for GPU acceleration. Recent
GPGPU studies showed experimental results of many image
processing algorithms in Table 1 when CPU and GPU are used,
respectively [4][11][12]. As shown in this table, for most of
image processing algorithms, GPU was more efficient than
central processing unit (CPU). In contrast, for the speeded-up
robust feature (SURF) detector, GPU did not perform as fast as
CPU. The SURF detector requires more complex computation
than the others in the table [5][6].
According to the complexity degree of image processing
algorithms, GPU is efficient or not in terms of the processing
speed. Moreover, in the AR applications, GPU should perform
the artificial information rendering. Therefore, if all image
processing algorithms for the object detection are assigned to
GPU, the over processing time of the AR application may not
decrease. Consequently, using both of CPU and GPU can
provide better performance than using only GPU and using
only CPU [7]-[10].
We present a parallel processing scheme using both CPU
and GPU in mobile devices. This scheme focuses on enhancing
the processing speed of the object detection.