摘要
搜救、安保等领域探测设备一般较为笨重,使用不便。提出了一种创新结构的小型模块化球形探测机器人系统,体积小巧、使用灵活,可通过磁力连接固定装置,将核心模块与机动模块快速组合,满足不同场景的探测任务需要。机器人系统通过鱼眼摄像头、高清摄像头等传感器进行环境感知,以树莓派及英特尔神经计算棒进行低功耗嵌入式计算,能够实现图像的实时处理和复杂场景下的多目标检测。通过原理样机实验,证明了所提出机械、电气设计和基于深度学习的目标识别方法的可行性和有效性。
Exploring robot are widely used in rescue,security and similar tasks.However,bulky appearances lead to inconveniences in their applications.In order to achieve light weight and flexibility,a minitype modularized exploring robot was proposed.The robot's system consisted of a core module and two driving modules,connected through a magnet-based fast assembly.Images from an onboard fish-eye camera and a high-resolution camera in real time were processed by an embedded low-energy-cost calculation system powered by Raspberry Pi and Intel Movidius Neural Compute Stick on the robot for instance,using single shot multiBox deteetor(SSD)model to perform multi-object detection.Experiments conducted on robot prototypes demonstrate the feasibility and effectiveness of the proposed mechanical and electronical design,as well as the deep learning based detection method.
作者
李勐
刘弋锋
袁柳
湛月
LI Meng;LIU Yi-feng;YUAN Liu;ZHAN Yue(China Academy of Electronics and Information Technology,Beijing 100041,China)
出处
《科学技术与工程》
北大核心
2019年第20期252-257,共6页
Science Technology and Engineering
基金
国家重点研发计划(2017YFC0821603)
中国电科联合基金(6141B0801010a)
博士后科学基金(2018M631542)资助
关键词
探测机器人
模块化
目标识别
鱼眼摄像头
exploring robot
modularized robot
object detection
fish-eye camera