Shadow detection and elimination is critical in traffic vision analysis, since shadow regions are often misclassified as object region. This leads to errors at segmentation stage and it results in poor tracking and a false classification of moving objects. This paper presents a novel shadow elimination approach that considers intensity properties through the histogram and texture features. First the segmentation of moving foregrounds is performed from background by using a background subtraction technique followed by K-means clustering. Then, we analyze the texture information and proximity in terms of similarity between a pair of segmented regions for determining shadow regions. At last step a Gaussian model is proposed to remove dynamic shadows. Experimental results validate the algorithm's performance.