| Counting and diameter measurement of the rebars are important processes in reinforced concrete building construction.Due to a large number of bundles of rebars and the diversity of diameter specifications,the traditional manual counting and diameter measurement methods will consume a lot of time and energy,which will easily lead to personnel fatigue and detection accuracy.With the rapid development of technologies such as smart construction sites,the construction industry is gradually introducing modern technologies such as automation,informatization,and intelligence to improve work efficiency.At present,with the rapid development of artificial intelligence,computer image processing methods such as target detection based on deep learning are more and more widely used,and their detection accuracy is also continuously improving.Based on machine vision technology,the computer can automatically count and measure the diameter of rebars,effectively deal with the complex construction site environment,reduce the burden on personnel,and improve the accuracy and efficiency of operations.This thesis focuses on the research of vision-based rebar counting and diameter measurement methods.The main contents are as follows:1.To improve construction efficiency and reduce the burden on workers,a counting and diameter measurement of the rebar system is constructed.By encapsulating the visual algorithm,and interacting with the mobile phone.The workers shoot the ends of the bundled rebars and photograph a single rebar at a fixed overlooking angle and upload it to the server,so as to realize the function of counting and diameter measurement of the rebars.2.To improve the accuracy of vision-based rebar counting,the anchor frame clustering,network structure,and loss function of convolutional neural network target detection algorithm Yolov4 are improved,and a lightweight model Steel-Yolov4 algorithm is proposed.The improved algorithm has stronger robustness in the face of large changes in the construction site environment,high density of bundled rebars,and occlusion due to uneven placement of rebars.3.To improve the accuracy of vision-based rebars diameter measurement,the rebars diameter measurement method combining convolutional neural network semantic segmentation algorithm and image processing is proposed for the defect of weak generalization performance of traditional image morphological processing methods in the face of background changes.Aiming at the problem that the rebars image captured by overlooking is covered by shadows and the outer diameter is small and not easy to be distinguished,a Steel-Encoder-Decoder lightweight semantic segmentation algorithm is proposed to segment the rebars image.The segmented rebar image is processed by image morphology and various numerical calculations to fit the diameter of the rebar.Experiments show that in the test set of rebar counting,the detection accuracy Map of the Steel-Yolov4 algorithm is 99.40 %,and the number of model parameters is only 1/4 of the original algorithm.In the test set of semantic segmentation of the rebars,the detection accuracy MIo U of the Steel-Encoder-Decoder semantic segmentation algorithm is 97.58 %,and the fitted diameter error does not exceed 0.5mm,which meets the rebars acceptance standard specification. |
