Light-invariant 3D object's pose estimation using color distance transform

This paper presents a novel pose estimation method based on color distance transform. Although the distance transform has already been used for object detection, pose estimation and active contouring we use it in a completely novel way to isolate the colors from which the pose is estimated. Standard matching and pose estimation techniques usually rely on the available texture and feature points. However, they are bound to fail in low texture or extreme noise presence conditions. The edge-based approaches can overcome such failures but they are generally too slow for real-time performance. Moreover the image luminance affects these techniques as the edges are not visible and in many cases the foreground and the background are merged. On the other hand, we obtain robustness against extreme lighting conditions by using the HSV-HSL color spaces. We isolate the colors operating only on the chromaticity plane where luminosity (L) and value (V) have no actual effect. Depending on the vectors' norms in the RGB cube we obtain the chroma which is subtracted from the HSV-HSL color disk leaving only the colors with high intensities. The distance transform is then applied on the remaining colors in order to find their `sources'. The pixels which have the same color but are far from the source are excluded, since they are considered noisy. The proposed method maintains its accuracy and precision against different lighting conditions without increasing the processing time or the computational resources.