Wheel overheat detection via incorporation of scattering brightness difference and texture representation

Wheel overheat is easy to cause spontaneous combustion, and makes severe traffic accidents. To this end, this paper contributes a novel wheel overheat detection system based on far-infrared video analysis. With effective frame selection of wheel video, we propose a scattering brightness difference descriptor (SBDD) to represent the brightness distribution of wheel infrared image. In addition, the traditional and superior texture features, such as histogram of oriented gradient (HOG) and local binary patterns (LBP), are integrated to make a more powerful representation of wheel image. The extracted low-dimensional feature set is projected into high-dimensional space by RBF kernel, and the support vector machine is utilized to judge the wheel overheat. To restrain the influence of image border and background, wheel feature region is selected by a Gaussian kernel. Experimental results show that the proposed system can accurately find the wheel overheat, and has important significance to reduce the traffic accidents.