未知环境有害气体烟羽跟踪问题研究

It is important to explore the law of environmental perception by applying harmful gas plume tracing method in unknown environment, because harmful gases not only have adverse effect on human life and the environment, but also often cause serious accident. This study will contribute to effective gas monitoring in working condition and provide reference for mobile robot which works in unknown environment. Firstly, the layered tracing idea is proposed for unknown environment. Secondly, the hierarchical multi-scale feature points are detected by SIFT algorithm, and the relationship between environmental segmentation and feature extraction is investigated so as to exposit the law of visual environmental perception. Thirdly, the depth fusion model of the olfactory information and the visual information is constructed, the influence factors are analyzed so as to improve the gas tracing efficiency and explore the law of olfactory and visual fusion environmental perception. Finally, the experiments are carried out by the layered plume tracing method, and the results are compared with the other data that are made by other tracing algorithms. Then, the depth fusion model is modified and the law of environmental perception is deduced.

有害气体不仅破坏生态环境、危害人体健康，还易引发爆炸、火灾等重特大事故，研究未知环境有害气体烟羽跟踪问题，探索未知环境感知规律，可为实际工况有害气体监测寻踪提供依据，为机器人在未知环境下作业提供指导，具有重要的科学与社会意义。首先，针对实际工况中湍流、风速、视觉等信息均未知的情况提出分层渐进跟踪思想；其次，基于SIFT算法进行分层多尺度特征点提取，通过研究特征点参数对特征分区的影响，揭示未知环境视觉感知规律；再次，为提高跟踪效率，创建嗅觉信息与视觉信息的深度融合模型，仿真分析各种因素对融合结果的影响，探索未知环境嗅视觉感知规律；最后，构建烟羽跟踪策略进行未知环境有害气体烟羽跟踪实验研究，通过分析不同跟踪策略的实验结果，总结各种因素对烟羽跟踪的影响，验证和完善嗅视觉信息深度融合模型，进而推理未知环境感知规律。

本课题针对未知环境有害气体烟羽跟踪问题进行研究。对气体传感系统进行了优化研究，降低了系统噪声，并将激光的可调谐范围从30nm拓宽至60nm；引入了自适应Kernel算法并对其进行改进，实现了去噪的同时较好的保留吸收谱线型；提出了凸点拟合算法，提高了基线提取的精度；创建了多尺度吸收峰检测模型及分段低阶多项式拟合模型，并提出了相应的基线提取算法，提高了吸收峰检测的精度和实时性；选用合适的吸收峰拟合算法及多吸收峰协同提高浓度解调的精度和稳定性；建立了全方位视觉系统模型，研究采用球面模型法和函数法进行畸变校正，通过函数法进行畸变校正提高了图像校正的效果和实时性；基于SIFT算法实现了分层环境特征点提取，借鉴SIFT算法的部分思想，结合统计分析，提出了基于显著度的特征提取算法，该算法特征点提取的实时性和准确性均较高，并基于该算法进行了斑点提取研究；研究了特征点强度阈值以及特征点个数对环境区域划分的影响，总结了以全局特征点强度为基础进行环境区域划分和以局部特征点集总强度为基础进行环境区域划分的优缺点；提出一种基于模糊控制的气体扩散模型选择的方法，用视觉传感器和嗅觉传感器共同获取环境信息，通过多源信息融合减少噪声和仪器故障的影响，提高疑似泄漏源判断的效率和烟羽跟踪效率；分析了气体浓度计算方法、气体浓度阈值设定、环境距离阈值设定等对烟羽跟踪结果的影响。