Learning to self-adaptive robust dehazing network for video detection in drainage pipeline

Abstract

With the continuous urbanization and expansion of the built-up area of cities, the requirement for regular maintenance of urban underground drainage pipeline will be further increased. However, the drainage pipeline environment is humid and the water quality conditions are complex, which leads to the existence of uneven concentration of haze in the images captured by CCTV (closed-circuit television). To address these problems, this paper suggests a self-adaptive robust dehazing network for video detection in drainage pipeline. The method extracts feature in the encoder by self-adaptive module and fully multi-scale feature alignment module, and the dehazed image is gradually recovered in the enhanced decoder after the feature recovery module. The results of the experiments indicate that this method significantly improves the robustness and generalization ability of model dehazing by capturing global hazy features at multi-scale resolution, and it is an effective method for applying to drainage pipeline dehazing.