【摘要】 视觉系统是蚱蜢仿生机器人必不可少的部分,因此研究视觉系统有重要的意义。从软件和硬件系统两个方面来介绍蚱蜢仿生机器人的视觉系统。在搭建蚱蜢仿生机器人视觉系统的硬件时,考虑到蚱蜢仿生机器人的空间小、结构紧凑、形态小巧,无法安装太多的硬件模块,因此对硬件系统进行了最简化的设计。其中视觉系统硬件是由两个摄像头、4g通讯模块路由模块以及数据传输线组成,以及为摄像头和4g传输模块供电的锂电池组。其工作原理是,摄像头对周围环境进行拍摄视频,4g模块将摄像头拍摄的视频传输到云服务器;软件系统是主要是有两部分组成:1)对视频进行处理,将对拍摄的视频进行抽帧转换成图像、在对图像进行进降噪的处理,为第二步目标检测做准备。2)进行目标检测。对处理好的图像的进行检测。在目标检测方面,是基于yolov3的深度神经网络算法,具有检测速度较快、准确率高。通过软硬件系统的设计搭建,能够满足蚱蜢仿生机器人实现对障碍物自主识别。通过实验验证识别率能达到95%以上。更多还原

【Abstract】 Vision system is an essential part of grasshopper bionic robot, so it is of great significance to study vi-sion system. The vision system of grasshopper bionic robot is introduced from two aspects of software and hardwaresystem. This article introduces the vision system of grasshopper bionic robot from two aspects of software and hardwaresystem. When constructing the hardware of the grasshopper bionic robot vision system, considering that the grasshop-per bionic robot has a small space, a compact structure, and a compact form, and cannot install too many hardwaremodules, the most simplified design of the hardware system was carried out. The vision system hardware consists oftwo cameras, a 4 G communication module routing module, routing module and data transmission line, as well as alithium battery pack that supplies power to the camera and 4 G transmission module. Its working principle is that thecamera takes video of the surrounding environment, and the 4 G module transmits the video taken by the camera to thecloud server. The software system was mainly composed of two parts: 1) processing the video, converting the cap-tured video into an image, and further reducing the noise of the image, so as to prepare for the second step of targetdetection; 2) Detecting the of the processed image. In the aspect of target detection, a deep neural network algorithmwas used based on yolov3, which has fast detection speed and high accuracy. Through the design and construction ofthe software and hardware system, the grasshopper bionic robot can realize the autonomous recognition of obstacles.Experiments show that the recognition rate can reach more than 95%.更多还原