Vehicle re-identification (V-ReID) is one of the major technologies of intelligent transportation and traffic management systems, which is an important aspect of the development of smart cities. The speedy developments in the field of Deep Learning (DL) methods, re-identification techniques achieved substantial improvement in recent years. But this research area received less attention from computer vision enthusiasts due to the shortage of proper datasets. Earlier studies were mainly focusing on specific views (i.e., front/back), but these techniques are not effective in real-world scenarios, where the angle and visual appearance of the moving vehicle change for the camera. In this article, we propose a dual-stream Siamese network for vehicle re-identification by substitution of few traditional convolutional layers into dilated convolutional layers (DCL). In our proposed framework, both streams are utilizing baseline ResNet-50 model, where one is employed to extract features from registration plates while the next stream capture features from vehicles shape. We trained our model on ten videos and tabulated the details about the proposed two-stream network. Experimental results indicate consistent improvement and accuracy.