Bilateral Filtering and Hybrid Homomorphic Normalization for Lane-Image Pre-processing
Mark Paulik, Ahlam Al Mohammad · 2024
In this work we introduce a novel approach for lane-image pre-processing that normalizes illumination, preserves lane lines and significantly reduces image noise. The method presented incorporates a new illumination correction algorithm based on dual-channel homomorphic filtering principles combined with grayscale morphology to achieve dynamic range reduction, local contrast enhancement and multiplicative noise reduction. Bilateral filtering is subsequently employed to significantly smooth the image while preserving lane edges. The resulting images are highly suitable for subsequent processing to extract and track roadway lane lines using either neural network or classical image processing techniques. An extensive experimental study demonstrates the effectiveness of the proposed technique.