煤炭传送带工作状态的巡检是保障安全生产的重要环节，巡检机器人作为智能化巡检常规手段，对于保障煤矿安全生产、实现井下减人增安具有重要意义。为了提高矿井巡检机器人在长距离、复杂工况下的避障能力，实时调整行径路线并实现对障碍物的避让，基于红外摄像仪与激光雷达多源传感融合的巡检避障系统，提出了一种基于子图像分割映射点云空间的机器人避障算法，首先通过将红外数据作为边界条件，将红外图像分块构成子图像单元，以不同尺度的子图像对点云空间范围进行映射，进而实现对障碍物点云进行提取；然后利用各个子图像单元进行投影的方式，完成对目标区域三维点云的限制；其次利用边界约束降低点云数据处理总量的方式，提高算法收敛速度和障碍物特征点云的提取速度；最后通过仿真分析验证了算法对点云总量的精简效果，模拟了障碍物在不同子图像尺度下的最大外径反演精度，验证了应用本算法的系统实时避障能力效果。试验结果表明：当子图像边长为10.0 mm时，最大相对误差低于1.53%，算法收敛时间为1.243 s，障碍物外径反演精度与收敛速度均达到实际应用需求；算法在静态障碍物、动态障碍物以及多机器人避障环境中均有较高的准确率和避障效率，满足巡检机器人实时进行环境数据采集与避障的需求，具有较高的实际应用价值。

The inspection of the working status of coal conveyor belts is an important part of ensuring safe production. As a conventional means of intelligent inspection, inspection robots are of great significance for ensuring the safe production of coal mines and realizing the reduction of personnel and the enhancement of safety underground. In order to improve the obstacle avoidance ability of mine inspection robots under long-distance and complex working conditions, adjust the travel route in real time and achieve the avoidance of obstacles, based on the inspection and obstacle avoidance system with the fusion of multi-source sensing of infrared cameras and lidars, the main work is as follows: A robot obstacle avoidance algorithm based on sub-image segmentation and mapping of point cloud space is proposed. Firstly, by taking infrared data as the boundary condition, the infrared image is divided into blocks to form sub-image units, and the point cloud space range is mapped with sub-images of different scales, thereby realizing the extraction of obstacle point clouds; and by using the projection method of each sub-image unit, the limitation of the three-dimensional point cloud in the target area is completed; then, by using the way of boundary constraints to reduce the total amount of point cloud data processing, the convergence speed of the algorithm and the extraction speed of the obstacle feature point clouds are improved. Finally, the simplification effect of the algorithm on the total amount of point clouds is verified through simulation analysis, the inversion accuracy of the maximum outer diameter of obstacles under different sub-image scales is simulated, and the real-time obstacle avoidance ability effect of the system applying this algorithm is verified. The experimental results show that when the side length of the sub-image is 10.0 mm, the maximum relative error is less than 1.53%, the convergence time of the algorithm is 1.243 s, and both the inversion accuracy of the obstacle outer diameter and the convergence speed meet the actual application requirements; the algorithm has a high accuracy rate and obstacle avoidance efficiency in static obstacle, dynamic obstacle and multi-robot obstacle avoidance environments, meets the needs of inspection robots for real-time environmental data collection and obstacle avoidance, and has a high practical application value.