|
Ultra fast cnn based Hardware Computing Platform Concepts for adas visual Sensors and Evolutionary Mobile RobotsBog'liq Alireza FasihFigure
8-4: OpenCL Programming Flow
In our work we have used the combination of OpenCV for the appropriate user interface
and for data acquisition, and OpenCL for the development of the desired UM-CNN on the
heterogeneous platform.
For real time image processing applications the performance of CNN is evaluated for the
contrast enhancement. The templates used for the contrast enhancement are given below
(8-7)
𝑇 =
0
0.25
0
0.25
0
0.25
0
0.25
0
; 𝑇 =
0
−1
0
−1
4
−1
0
−1
0
; 𝐼 = 0
94
Figure 8-5 shows the given input image and output image of the CNN operation on GPU
and Fig 8-6 shows the output image of the CNN operation on CPU.
Figure
8-5: (a) Input image for CNN (b) Output image of CNN on GPU
Figure
8-6: Output of CNN on CPU after applying enhancement template
8.7
Conclusion
In this work we propose a new model for image processing that can be executed in parallel
and faster. We implement this model with OpenCL framework along with OpenCV. The
concept of DT-CNN on the GPU makes the execution of the image processing task faster.
The concept of high performance computing is achieved with the designed model. Till now
the discussion goes on with the surrounding knowledge of the designed model. In the
95
future work, a clear description of the entire model along with the benchmarks of
comparing the performance of GPU with performance of CPU will be done.
|
| |
