Image enhancement for UAV visual SLAM applications: analysis and evaluation

Although simultaneous localisation and mapping (SLAM) has been widely applied in a wide range of robotics and navigation applications, its applicability is severely affected by the quality of the acquired images, especially for those in unmanned aerial vehicles (UAV). In this paper, comprehensive analysis and evaluation of the methods for enhancement of the UAV images are focused, especially the models for denoising of the UAV images using spatial-domain analysis, transform domain analysis and deep learning. Experiments on publicly available datasets are conducted for performance evaluation, along with both qualitative and quantitative results. Surprisingly, deep learning-based approaches did not perform particularly well as these did in other computer vision tasks such as object detection and recognition. Useful discussions are suggested how to further explore this interesting topic.