Defect Detection of the Irregular Turbine Blades Based on Edge Pixel Direction Information

In this paper, we proposed a new method for detecting the defect parts of irregular turbine blades in real-time based on edge pixel direction information. Firstly we detect and track the turbine target with the background subtraction algorithm. And extract possible defective blades in fixed positions of the turbine. Then we split the whole system into two parts: 1) detecting the defect on the left/right terminal of turbine blade, where the well-known statistical HOG feature descriptor is applied to the extracted image corner. The defect blade could be found if its HOG feature is significantly different from that of a normal one; 2) detecting the defect in the middle blade with a new method where the shape changing of the blade edges caused by defect will be determined by applying the least-square method to fit the direction vectors within an adaptive detection window. The experimental results show the effectiveness of the proposed methods. The processing speed is about 11 fps, which basically meets the real-time requirements.