Offloading Image Recognition Processing Using Heterogeneous FPGAs for Care Robot Applications

• Item type: Symposium(1)
• 公開日: 2023-12-20
• タイトル: Offloading Image Recognition Processing Using Heterogeneous FPGAs for Care Robot Applications
• タイトル / 言語: en
• タイトル: Offloading Image Recognition Processing Using Heterogeneous FPGAs for Care Robot Applications
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_5794
• 資源タイプ: conference paper
• 著者所属: Tokai University
• 著者所属: Shibaura Institute of Technology
• 著者所属: Tokai University
• 著者所属: Tokai University
• 著者所属: Shibaura Institute of Technology
• 著者所属: Kumamoto University
• 著者所属(英): en
• 著者所属(英): Tokai University
• 著者所属(英): en
• 著者所属(英): Shibaura Institute of Technology
• 著者所属(英): en
• 著者所属(英): Tokai University
• 著者所属(英): en
• 著者所属(英): Tokai University
• 著者所属(英): en
• 著者所属(英): Shibaura Institute of Technology
• 著者所属(英): en
• 著者所属(英): Kumamoto University
• 著者名: Eisuke, Okazaki; Hayato, Mori; Gai, Nagahashi; Mikiko, Sato; Midori, Sugaya; Takeshi, Ohkawa
• 著者名(英): Eisuke, Okazaki; Hayato, Mori; Gai, Nagahashi; Mikiko, Sato; Midori, Sugaya; Takeshi, Ohkawa
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Nursing care robots are expected to reduce caregiver labor by patrolling the care facility and detecting falls of older adults. However, the image recognition processing of robots has performance and power consumption issues. To solve these problems, our approach is to use heterogeneous FPGAs as the computational resource of the edge server. This paper shows the concept of offloading image recognition processing from the robot to the edge server. As an evaluation, the throughput of pose estimation and face detection was measured using heterogeneous FPGAs. The results show that throughput improvements of 8.71 times and 119 times in face detection and pose estimation, respectively, can be achieved compared to the Raspberry Pi 4B. In addition, results show that the use of heterogeneous FPGAs is advantageous because the FPGA with the DPU that achieves the highest throughput depends on the type of image recognition processing.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Nursing care robots are expected to reduce caregiver labor by patrolling the care facility and detecting falls of older adults. However, the image recognition processing of robots has performance and power consumption issues. To solve these problems, our approach is to use heterogeneous FPGAs as the computational resource of the edge server. This paper shows the concept of offloading image recognition processing from the robot to the edge server. As an evaluation, the throughput of pose estimation and face detection was measured using heterogeneous FPGAs. The results show that throughput improvements of 8.71 times and 119 times in face detection and pose estimation, respectively, can be achieved compared to the Raspberry Pi 4B. In addition, results show that the use of heterogeneous FPGAs is advantageous because the FPGA with the DPU that achieves the highest throughput depends on the type of image recognition processing.
• 書誌情報: Proceedings of Asia Pacific Conference on Robot IoT System Development and Platform; 巻 2023, p. 73-74, 発行日 2023-12-20
• 出版者 / 言語: ja
• 出版者: 情報処理学会