Real-time detection method and device for different types of entity objects in construction site image

Abstract

The invention discloses a machine vision based real-time detection method and device for different types of entity objects in a construction site image. The method comprises the steps that 1) a trained convolutional neural network is used to extract a characteristic image from an original image of the construction site; 2) an area suggestion network is used to extract a candidate area from the characteristic image; 3) the candidate area is used in the rapid convolutional neural network; 4) the area suggestion network and the rapid convolutional neural network share a convolution layer to obtain a real-time convolutional neural network; and 5) the real-time convolutional neural network is used to identity entity objects in the to-be-detected construction site image. The device comprises computer programs to execute the method. Thus, different types of entity objects in the construction site can be identified in real time, and significant basis is provided for progress and safety management in the whole process of building engineering construction in the complex environment.

Description

technical field [0001] The invention belongs to the field of construction engineering informatization, and more specifically relates to a method and equipment for real-time detection of multiple types of entity objects in construction site images. Background technique [0002] The dynamics and complexity of the construction environment lead to a high number of accidents and fatalities during the construction period, and unsafe behavior and unsafe state are the two direct causes of accidents. If unsafe behaviors and states in construction can be monitored and understood in real time, safety performance will undoubtedly be improved. Real-time identification of construction objects such as people, machinery, and materials involved in construction operations in construction site video surveillance is critical to fully understanding unsafe behaviors at complex construction sites. At the same time, object detection in images / videos is also the basis for productivity measurement a...

AI Technical Summary

Problems solved by technology

[0004] (1) It is obviously not enough for complex construction sites involving many types of machinery, materials and other construction objects;

[0005] (2) Due to the lack of comprehensive detection of the entire image area, specific applications will be greatly limited

[0006] Since the current research on identifying more than two types of construction objects is still rare, and when it is applied to the detection of equipment and workers, it is limited to the traditional paradigm of machine vision, and the effectiveness and comprehensiveness of feature extraction are greatly limited. Model generalization Poor ability. In addition, because existing methods rely on manual feature extraction and cannot perform computationally intensive processing, real-time recognition cannot be achieved

Images