Automatic Detection of Transmission Line on UAV Inspection Images with the Statistics Approach in the DCT Domain

Power transmission lines are significant aspect for uninterrupted electric energy supply. The aerial camera captured image processing methods use for inspecting the condition of this infrastructure involve algorithms that explicitly find line segments in the image and classify them according to width, length and angle. These methods are erroneous with extreme false detection outcome rate. It’s obvious that image may contain a complex background with a line like structures. For this purpose, an alternative approach that visualizes the statistical model of local DCT coefficients is proposed to estimate the existence of power line. The proposed algorithm tackles the issue of power transmission line detection by exploiting low-level images characteristic to mimic human visual acuity and detect the existence of power line by mapping the extracted coefficient to Generalized Gaussian Distributions (GGDs) from the DCT coefficients. Then percentile pooling is applied on the fitted coefficient to obtain values representing the local and global distribution. Emphasis is laid on the orientation feature to capture the transmission line varying position and location in separate images. These features are feed as input to SVM for classification. The proposed algorithm was tested on the TEIAS visible light database containing 4000 undistorted images, and a local dataset containing 1039 distorted and undistorted images. Experimental results show the proposed method can detect power lines with reasonable performance.