This paper proposes to improve the approach presented in Krähenbühl et al. [11] to build automatic methods for the wood knot detection from X-Ray CT scanner images. The major drawbacks of the previous method mostly depends on the variety of the distribution of knots and their geometric shapes. Our aim is to extend the robustness by performing the accumulation process of Z-Motion differently and by suppressing the whorl distribution constraint. This is achieved both through a polar Z-Motion accumulation and an aggregation process of connected components related to maxima localization in the accumulation space. The experimental results are in favor of an increase in the robustness while being more sensitive to small and isolated knots. This opens the way to a method fully independent of wood species.