¿µ¹®³»¿ë (English Abstract) |
It is difficult to implement the distributed multicamera object recognition system due to the limited computing resource of edge devices and network conditions. This paper includes specification of the implemented system, with the solutions to the problems raised during implementation. By utilizing TensorRT, edge selection and network latency hiding, our model achieved 20 FPS speed and 77% accuracy, which is 1.52x faster in terms of FPS and 18% improvement of accuracy, compared with the baseline model. |