A Joint Faster RCNN and Stereovision Algorithm for Vegetation Encroachment Detection in Power Line Corridors

Abstract

With the rapidly increasing construction of power lines in complex terrains, vegetation encroachment of overhead power lines can cause severe consequences, e.g., wildfires and cascading failures. Therefore, it is significant to detect vegetation encroachment of overhead power lines to ensure secure and reliable operation of power systems. This paper proposes an automatic joint algorithm to fast and accurately detect vegetation encroachment of overhead power lines in real-time. This algorithm integrates a deep learning method, i.e., the region convolution neural network (Faster RCNN) into a stereovision algorithm to process the 2D image data captured by the vision sensors located on a power tower. The generated bounding boxes (bboxes) only can represent vegetation regions in 2D images that contain the features of vegetation. The stereovision algorithm is applied in this paper to process the bboxes in 2D images to generate the corresponding 3D information for vegetation location, including the height of the vegetation. The proposed algorithm has been tested on the realistic vegetation encroachment image data from an utility. The results show that the proposed algorithm can fast and accurately detect vegetation encroachment of overhead power lines.