基于视觉的巡检机器人环境感知和导航避障研究

Research on Environment Perception and Navigation ObstacleAvoidance of Inspection Robot Based on Vision

为提升巡检机器人环境感知和导航避障性能,将深度学习技术与路径规划算法相结合,提出了一种基于卷积神经网络和改进RRT算法的视觉避障方法。该方法主要分为环境感知和路径规划两部分,环境感知采用编码-解码架构,以深度可分离卷积、H-Swish激活函数结合注意力机制、空间金字塔采样等结构实现图像像素级分类,保障巡检机器人充分感知所处环境;路径规划则根据所感知的环境构建栅格地图,并设计基于启发式的RRT路径规划算法搜索出安全有效的避障路径。为提升算法在实际应用中的效率,引入了自主控制机制,通过计算相邻视频帧的信息差异来自主调节网络的数据流向和路径规划的候选节点,减少冗余计算。实验结果表明,所提方法有效提升了机器人环境感知能力,并能准确规划出安全的导航避障路径。在实际电网场景中,该方法也体现出较高的适应性,可以高效辅助巡检机器人完成巡检任务。

In order to improve the environment perception and navigation obstacle avoidance performance of inspection robot,a visual obstacle avoidance method based on convolutional neural network and improved RRT algorithm is proposed by combining deep learning technology with path planning algorithm.The method is mainly divided into two parts:environment perception and path planning.The environment perception adopts the encoding-decoding architecture,and uses the depth separable convolution,H-Swish activation function,attention mechanism,spatial pyramid sampling and other structures to achieve pixel-level classification of image,ensuring that the inspection robot can fully perceive the environment.Path planning builds a grid map according to the perceived environment,and designs a heuristic RRT path planning algorithm to search safe and effective obstacle avoidance paths.At the same time,in order to improve the efficiency of the algorithm in the practical application,an autonomous control mechanism is introduced to reduce redundant computation.The data flow direction of the perceptual network and the candidate nodes of the path planning are autonomously adjusted by calculating the information difference of adjacent frames.The experimental results show that the proposed method effectively improves the robot s environmental awareness,and can accurately plan a safe navigation obstacle avoidance path.In the actual power grid environment,the method also shows high adaptability and can efficiently assist the inspection robot to complete the inspection task.