The relay protection operation requires strict technical operation and high accuracy. In response to the current challenges of high complexity and poor universality in modeling with augmented reality (AR) applications, a feature extraction edge computing framework is proposed. Firstly, the Canny algorithm is applied to perform edge analysis on the meter, pressure plate, and terminal block of the protection cabinet. To address salt-and-pepper noise in the images, a fixed-window median filtering method is employed for prepossessing. The improved Sobel operator optimizes gradients in the horizontal and vertical directions, and a set of contour boundary lines is constructed using the least squares method to handle points with significant gradient changes. Subsequently, the Harris algorithm is utilized for corner point analysis, and an improved corner point response function with fixed parameters efficiently determines corner points within the contour boundary line set. Finally, object recognition on the protection cabinet is carried out based on templates. Experimental results demonstrate that the proposed algorithm exhibits low computational complexity, fast response speed, and high accuracy, meeting the requirements for on-site operations.