| As the support system of building structure,the process quality control of concrete project is the lifeline of construction enterprise.As the process of building modern infrastructure system advances,the traditional way of concrete construction control has high labor cost and low efficiency,and it is urgent to upgrade and transform to intelligent management.According to the current research status of object detection and action recognition algorithms,while considering environmental issues at the construction site,this topic proposes a real-time object detection algorithm for complex construction environments and a action recognition algorithm that incorporates motion capture and spatio-temporal attention,and uses the above algorithms to build an efficient and low-cost concrete construction process quality monitoring system,which includes:(1)A real-time object detection algorithm for complex construction environments is proposed for the problems of cluttered environment,obscured objects,large object scale range,unbalanced positive and negative samples,and insufficient real-time performance of existing algorithms in the practical application scenarios of this project.The low and middle layer feature maps are input into the feature fusion network,and a new small object detection layer is added to improve the detection accuracy of the model for objects of different scales;the channel-spatial attention module is designed and added to the CSP structure so that the model emphasizes object features to suppress background features;in the loss function part,VariFocal Loss is used to calculate the classification loss of the model to solve the problem of unbalanced positive and negative samples;the GhostConv is used as the basic convolutional block to replace the original convolutional block and build the GCSP structure to reduce the number of model parameters and computation;construct the construction site concrete water addition detection dataset,and the experiments show that the proposed algorithm has improved accuracy and speed compared with YOLOv5s.(2)Aiming at the problems of large computation and memory consumption for capturing local motion by optical flow estimation,large number of parameters for acquiring temporal semantic information by 3D convolution operation,and difficulty in extracting global semantic features of long videos while maintaining low computing power,a action recognition algorithm integrating motion capture and temporal attention has proposed.The deformable convolution is used to build a motion capture module to capture local motion details and achieve efficient modeling of local motion information;the temporal attention and spatio-temporal attention modules based on matrix decomposition are built to model global temporal information and make the model focus on key frames and their key regions;a video dataset of concrete water addition violations is constructed,and extensive experiments are conducted on public datasets UCF101,HMDB51,SSV2 and self-built datasets to verify the effectiveness of the proposed algorithm.(3)For the problems of high cost and low efficiency of traditional concrete construction control methods,a deep learning based concrete construction process quality monitoring system is designed and implemented.Based on the above proposed object detection and action recognition algorithm to achieve the water addition action recognition function,while the introduction of camouflage object segmentation and human key point detection algorithm to achieve the vibrating construction point mapping function,and the development of cloud-based management system and cell phone APP to provide construction site managers to efficiently control the quality of concrete construction.The system was successfully applied to several construction sites of CSCEC Northwest Construction Co.,Ltd.to effectively improve its construction management efficiency and significantly reduce management costs. |
