An adaptive recognition of abnormal behaviour in deep excavation support construction site of high-rise buildings

To address the problems of high target false acceptance rates, low accuracy in abnormal behaviour recognition, and lengthy recognition times in traditional methods, this study proposes an adaptive recognition approach for abnormal behaviour in deep excavation support construction sites of high-rise buildings. Key frames are extracted from surveillance videos using the fractional Fourier transform, and object detection is implemented with spatiotemporal graph convolutional network models. Based on the target detection results, a CNN-LSTM model is used to achieve adaptive recognition of abnormal behaviour by capturing the temporal and spatial features of the target. Experimental results show that the proposed method achieves a minimum target false acceptance rate of 2.43%, a maximum recognition accuracy of 99.12%, and a minimum processing time of 0.19 s.