Method and device for detecting helmets and work clothes in transformer substation

Abstract

The invention discloses a method and device for detecting helmets and work clothes in a transformer substation. The method comprises the steps: collecting an inspection image of transformer substation workers, and marking helmets and work clothes in the image; taking the marked image as the input of a YOLO3 model, forming a result sequence by output results, and storing the result sequence in a container A; performing result frame resetting calculation, label rematching and confidence rematching on repeated result frames, keeping other parts unchanged, acquiring a new result sequence, and storing the new result sequence in a container C; with the original sequence in the container A as an input sequence and the new result sequence in the container C as a target sequence, training a seq2seq model to obtain a result box conversion model; subjecting the YOLO3 model and the result box conversion model to connection in series and integration to obtain a fusion model; and inputting a to-be-recognized image into the fusion model, and deleting a sequence with a confidence coefficient smaller than a preset threshold value in the output result sequence to obtain a detection result without repeated frames. According to the invention, result frame positioning precision can be improved, and redundant frames are removed.

Description

technical field [0001] The invention belongs to the field of intelligent patrol inspection of substations, and particularly relates to a detection method for helmets and work clothes of substations. Background technique [0002] With the continuous development of computer vision, intelligent inspection has gradually become a key research field of substations, which involves detecting whether workers are wearing helmets, work clothes and other safety equipment correctly. Whether the work clothes are properly worn. [0003] At present, in target detection, a series of algorithms such as R-CNN and YOLO are the mainstream algorithms. Although the above algorithms can quickly and accurately locate the detection target, there is still a problem that the detection results will be repeated, that is, a target may appear multiple The result box, using the traditional non-maximum suppression processing method may cause problems such as missed detection or inaccurate positioning due to...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the present invention proposes to provide a substation helmet and work clothes detection method, device, equipment, and storage medium, which can solve the problems of repeated result frames and inaccurate positioning in existing substation target detection

Images