视觉神经网络林场巡检机器人的设计与开发

Design and Development of Forest Farm Inspection Robot Using Visual Neural Network

以传统移动机器人为基础,结合人工智能技术、传感器技术,设计技术方案——采用无刷直流电机(BLDC)构建1套3电系统,作为自动驾驶技术的研究平台;使用“感知-决策-控制”方式作为自动驾驶平台架构,应用单目相机作为视觉传感器;自建道路行驶图像与指示标志目标检测数据集。依据林场巡检的实际需要及设计的技术路线,设计了一款以神经网络技术为基础,使用单目相机作为传感器的自动视觉引导驾驶的林场巡检机器人;对设计的样机,在林场不同特性的不同环境道路中进行了实际行驶测试。结果表明:依据实际道路测试结果,视觉引导的可移动机器人达到了设计目标。确定的机器驱动执行机构方案与转向结构方案可行;为移动机器人平台设计构建的无刷直流电动机、电控系统、电池模组均工作正常,完成了机器平台的构建,实现了设计目标功能。依据“感知-决策-控制”的端到端方案在实际环境中的测试效果,证明使用简单传感器配合神经网络的视觉方案用于移动机器人自动驾驶中是可行的。在不同特性的不同环境道路(主要包括简易环形赛道环境、社区内部道路环境、林场环境、公路环境)、不同光照条件行驶测试中,均实现了自动驾驶功能,具有较高的灵活度;与当前主流使用的多传感器融合自动驾驶方案的移动机器人相比,采用单目相机作为传感器的端到端方案,无需使用繁杂昂贵的传感器便可实现其部分功能,成本大幅降低,自主运行移动机器人更简易化。

Based on the traditional mobile robot,combined with artificial intelligence technology and sensor technology,the design of technical scheme of brushless DC motor(BLDC)is used to build a set of 3-electric system as the research platform of autonomous driving technology.The“perception-decision-control”mode is used as the architecture of the autonomous driving platform,and the monocular camera is used as the visual sensor.Self-built road driving image and indicator target detection data set.According to the actual needs of forest farm inspection and the technical route of design,a forest farm inspection robot based on neural network technology and using monocular camera as sensor is designed.The designed prototype was tested on different environmental roads with different characteristics of forest farm.The visually guided mobile robot achieves the design goal according to the actual road test results.The determined machine drive actuator scheme and steering structure scheme are feasible.The brushless DC motor,electronic control system and battery module designed and constructed for the mobile robot platform all work normally,completing the construction of the machine platform and achieving the designed functions.According to the test results of the“perception-decision-control”end-to-end scheme in the actual environment,it is proved that the vision scheme using simple sensors and neural networks is feasible for the autonomous driving of mobile robots.In different environmental roads with different characteristics(mainly including simple circuit environment,road environment within the community,forest farm environment,highway environment),different light conditions driving test,automatic driving function is realized,has a high degree of flexibility.Compared with the current mainstream multi-sensor fusion autonomous driving scheme of mobile robots,the end-to-end scheme using a monocular camera as a sensor can realize some of its functions without the use of complicated and expensive sensors,the cost is greatly reduced,and the autonomous operation of mobile robots is easier.