摘要
针对煤矿机器人井下障碍物识别及位置测量精度较低的复杂问题,提出了一种基于改进Disp-NetC算法的模型。通过结合多尺度特征提取与注意力机制,增强图像特征的提取能力与匹配精度,旨在实现煤矿井下的障碍物识别与位置测量,进一步提高煤矿机器人智能化水平。实验结果显示,在大型数据集的训练过程中,模型的平均视差精度达84.05%,进一步证明了模型在复杂环境下的有效性。模型在VOC数据集和COCO数据集上的定位时间为157 ms和163 ms,定位精度为86.8%和87.1%,在精确度和定位速度上都具有明显优势。结果表明,研究提出的智能化定位技术是一种准确性和鲁棒性更高的定位方法,可以有效保障机器人的作业安全。
To address the complex issue of low accuracy in obstacle recognition and position measurement for underground coal mine robots,a model based on an improved Disp-NetC algorithm is proposed.By combining multi-scale feature extraction and attention mechanisms,the ability to extract image features and matching accuracy are enhanced,aiming to achieve obstacle recognition and position measurement in coal mines and further improve the intelligence level of coal mine robots.The experimental results showed that during the training process on a large dataset,the average disparity accuracy of the model reached 84.05%,further demonstrating the effectiveness of the model in complex environments.The localization time of the model on the VOC dataset and COCO dataset is 157ms and 163ms,with localization accuracies of 86.8%and 87.1%,respectively,demonstrating significant advantages in both accuracy and speed The results indicate that the intelligent positioning technology proposed in the study is a positioning method with higher accuracy and robustness,which can effectively ensure the safety of robot operations.
作者
李晓围
杨艳林
王皓雨
王再军
常健
LI Xiaowei;YANG Yanlin;WANG Haoyu;WANG Zaijun;CHANG Jian(Shendong Coal Group Co.,Ltd.,CHN Energy Buertai Coal Mine,Ordos,Inner Mongolia 017100,China;China coal Robot Technology Co.,Ltd,Shenzhen,Guangdong 518000,China;China coal(Liaoning)embodied intelligent technology Co.,Ltd.,Shenyang,Liaoning 110172,China)
出处
《中国煤炭》
北大核心
2025年第12期173-183,共11页
China Coal
基金
国家重点研发计划(2022YFB4703600)
辽宁省重大科技专项(2023JH26/10100006)。
