Context:
Embedded platforms GPUs are reaching a level of performance comparable to desktop hardware. Therefore it becomes interesting to apply Computer Vision techniques to modern smartphones. The platform holds different challenges, as energy use and heat generation can be an issue depending on load distribution on the device.
Objectives:
We evaluate the viability of a feature detector and descriptor on the Xperia Z3. Specifically we evaluate the the pair based on real-time execution, heat generation and performance.
Methods:
We implement the feature detection and feature descriptor pair Harris-Hessian/FREAK for GPU execution using OpenCL, focusing on embedded platforms. We then study the heat generation of the application, its execution time and compare our method to two other methods, FAST/BRISK and ORB, to evaluate the vision performance.
Results:
Execution time data for the Xperia Z3 and desktop GeForce GTX660 is presented. Run time temperature values for a run of nearly an hour are presented with correlating CPU and GPU activity. Images containing comparison data for BRISK, ORB and Harris-Hessian/FREAK is shown with performance data and discussion around notable aspects.
Conclusion:
Execution times on Xperia Z3 is deemed insufficient for real-time applications while desktop execution shows that there is future potential. Heat generation is not a problem for the implementation. Implementation improvements are discussed to great length for future work. Performance comparisons of Harris-Hessian/FREAK suggest that the solution is very vulnerable to rotation, but superior in scale variant images. Generally appears suitable for near duplicate comparisons, delivering much greater number of keypoints. Finally, insight to OpenCL application development on Android is given.
Source: Blekinge Institute of Technology
Authors: Danielsson, Max | Sievert, Thomas