Object Detection In Power Operation Scene Based On Two-Stream Network

In power production and construction, power accidents caused by the nonstandard operation of workers will not only bring economic losses but also cause personal injury. To supervise the non-standard operation of workers, this paper proposes an object detection method based on a two-stream Faster R-CNN. The main detection in the operation, whether the power workers in accordance with the requirements of electrical safety operation regulations, correctly wear a safety helmet and uniform, in order to ensure the safety of workers can get the maximum protection. According to the outstanding color feature of workers’ uniforms and safety helmets. In this paper, a two-stream network based on Faster R-CNN is proposed, in which the first-stream is RGB stream and the second-stream is HSV stream. The purpose of introducing HSV stream is to highlight the HSV color space features of the detected object, so that the final neural network model can be more sensitive to the HSV color space features of the detected object, so as to solve the false alarm problem when using Faster R-CNN model only. Compared with the Faster R-CNN method, the accuracy of the proposed method is improved by six percentage points.