The image enhancement of forward vehicle video based on catenary semantics

Abstract

In view of the complexity of existing image enhancement methods, which can not highlight the detection target, this paper proposes a forward vehicle image enhancement method based on the semantics. This method includes two parts: semantic-based catenary contour extraction and visual enhancement. To extract the catenary semantics contour, the classification model of the catenary and the background patches is trained by the catenary edge detection network. The classification model of the railway image is extracted and classified by the trained classification model, and the confidence level graph of the catenary is obtained by template matching. To achieve visual enhancement of catenary semantics, according to the AlphaBend hybrid method, the catenary confidence level graph is merged with the original image to realize the visual enhancement of the catenary semantics. In this paper, the contour extraction and visual enhancement of the catenary in the railway image are realized, and the gray histogram distribution is more evenly distributed. The number of pixels in each gray level is more average. The average difference and the standard deviation difference between the enhanced catenary area and the background area is greater, and the peak noise ratio and the structural similarity are improved. Comparing with other methods shows that the method in this paper is effective. What's more, it can be more intuitive for the railway staff to show the abnormal situation of the catenary and has very strong practical significance.