基于计算机视觉的太阳能光伏板轮廓提取与定位方法

Solar panel contour extraction and location method based on computer vision

针对车载机械臂清洁太阳能光伏板时需人工调整机械臂位姿的问题,提出利用计算机视觉提取太阳能光伏板外轮廓,并结合深度测量模型计算太阳能光伏板姿态信息的方法。搭建基于ResNet50网络的灰尘检测器,并改进下采样方法和引入注意力机制模块以提高检测精度。对于检测到灰尘的太阳能光伏板,将VoVNet27-slim网络作为DeepLabV3+的主干网络,采用GSConv卷积减轻模型的复杂度,同时将Decoder结构中的卷积模块改为MBConv以提高检测速率。在相似三角形测距算法的基础上,根据实际工况提出基于相机中心投影原理。结果表明,改进后模型的性能优于ResNet50和DeepLabV3+模型,在自动化生产方面具有较大的实用价值。

We proposed a method to extract the outer contour of solar photovoltaic panels using computer vision and calculate the solar photovoltaic panel posture information by combining with depth measurement model,aiming at the problem of manual adjustment of robot arm position when cleaning solar photovoltaic panels.A dust detector based on ResNet50 network was built,the downsampling method was improved,and an attention mechanism module was introduced to improve the detection accuracy.For the solar photovoltaic panels with detected dust,VoVNet27-slim was used as the backbone network of DeepLabV3+,and GSConv convolution was used to reduce the complexity of the model,while the convolution module in the Decoder structure was changed to MBConv to improve the detection rate.On the basis of the similar triangle-ranging algorithm,a depth measurement model based on the center projection principle of the camera was proposed according to the actual working conditions.The results show that the improved model performs better than ResNet50 and DeepLabV3+models,and has great practical value in automated production.