| Title |
Avoiding Performance Degradation of Object Detection via System Resource Management |
| DOI |
https://doi.org/10.5573/ieie.2020.57.1.77 |
| Keywords |
Deep neural network ; Object detection ; System resource management ; Mixed-criticality system ; |
| Abstract |
Recently, due to the rapid development of deep learning technology, the use of deep neural networks in the object detection of autonomous vehicles is becoming prevalent. Besides, the complexity and data requirements of deep neural network operations are rapidly increasing to show high recognition accuracy. When deep neural network computation is performed in cloud server, response characteristics due to network delay is inferior or there is a possibility of security problem. Recent advances in high-performance, multicore-based System on Chip (SoC) technologies enable neural networks to be computed in embedded environments. However, in this environment, system resources such as CPU and memory are shared with other applications, which can cause performance degradation when safety-sensitive functions such as object detection are performed. In this paper, we propose an operating system level solution that finds all tasks related to object detection, gives more system resource utilization if the performance interference occurs after the object detection is started, and returns to the original state when finished. The proposed solution is running on top of Jetson AGX Xavier, and the experimental results show that the performance of object detection is improved by 13.5%∼33% compared to before, under the performance interference of SPEC CPU2017 benchmark programs. |