Luo Xi, Feng Lei, Xun Hao, Zhang Yuanfei, Li Yixin, Yin Lihua
Cyber Space Institute of Advanced Technology, Guangzhou University, Guangzhou, China.
School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China.
Front Neurorobot. 2021 Aug 23;15:648101. doi: 10.3389/fnbot.2021.648101. eCollection 2021.
Image processing is widely used in intelligent robots, significantly improving the surveillance capabilities of smart buildings, industrial parks, and border ports. However, relying on the camera installed in a single robot is not enough since it only provides a narrow field of view as well as limited processing performance. Specially, a target person such as the suspect may appear anywhere and tracking the suspect in such a large-scale scene requires cooperation between fixed cameras and patrol robots. This induces a significant surge in demand for data, computing resources, as well as networking infrastructures. In this work, we develop a scalable architecture to optimize image processing efficacy and response rate for visual ability. In this architecture, the lightweight pre-process and object detection functions are deployed on the gateway-side to minimize the bandwidth consumption. Cloud-side servers receive solely the recognized data rather than entire image or video streams to identify specific suspect. Then the cloud-side sends the information to the robot, and the robot completes the corresponding tracking task. All these functions are implemented and orchestrated based on micro-service architecture to improve the flexibility. We implement a prototype system, called , and evaluate it in an in-lab testing environment. The result shows that is able to improve the effectiveness and efficacy of image processing.
图像处理在智能机器人中被广泛应用，显著提升了智能建筑、工业园区和边境口岸的监控能力。然而，仅依靠单个机器人所安装的摄像头是不够的，因为其提供的视野狭窄且处理性能有限。特别地，诸如嫌疑人等目标人物可能出现在任何地方，而在如此大规模的场景中追踪嫌疑人需要固定摄像头和巡逻机器人之间的协作。这导致对数据、计算资源以及网络基础设施的需求大幅激增。在这项工作中，我们开发了一种可扩展架构，以优化用于视觉能力的图像处理效率和响应速度。在该架构中，轻量级预处理和目标检测功能部署在网关端，以最小化带宽消耗。云端服务器仅接收识别后的数据而非完整的图像或视频流来识别特定嫌疑人。然后云端将信息发送给机器人，机器人完成相应的追踪任务。所有这些功能基于微服务架构实现和编排，以提高灵活性。我们实现了一个名为 的原型系统，并在实验室测试环境中对其进行评估。结果表明 能够提高图像处理的有效性和效率。
