Image Preprocessing Method for Visual Localization in γ Radioactive Environments

In γ radioactive environments, high-energy photons induce radiation noise in the camera image, which will seriously influence feature detection and tracking in the visual localization method. To mitigate these influences and avoid significantly increasing the computational complexity, in this paper, we propose a lightweight image preprocessing method. Firstly, the multi-layer neighborhood down-sampling strategy is used to split the speckle radiation noise. Then, the identified median filter is used to filter out the local flicker radiation noise in the sub-images and the neighborhood up-sampled reconstructed images. Finally, the contrast limited histogram equalization method is applied to improve the contrast of the image. The proposed method is verified by the scene images collected by a complementary metal oxide semiconductor camera exposed to radioactive sources in the Cobalt-60 hot cell. Experimental results show that compared with various image preprocessing methods, our method can effectively remove the radiation noise and significantly improve the feature matching rate.