<p>Intelligent traffic surveillance (ITS) systems are the essential need of any smart city. Such system monitors traffic, activities of vehicles and perform classification of detected vehicles types and categorizations of vehicles into related groups in real-time using the advanced computer vision and machine learning techniques. However, activities like classification of vehicles in low resolution and night vision video streams still a challenging task of ITS to deal with consistent efficiency. This work proposes a vehicle classification approach for low-resolution and night-vision images. The major aim of the proposed work is to utilize the advanced deep learning algorithms to enhance the capabilities of ITS. The framework proposed in this paper utilizes different deep learning models at the different levels to improve the efficiency of vehicle classification. The Approach specifically target the low resolution and night vision images for the classification of the vehicle in five different classes. The proposed approach is tested on a variety of the images such as image quality, different image viewpoint, pose and illumination. The proposed work achieves 96% average accuracy in the classification of vehicles during the testing. The proposed deep learning model is also compared with state-of-the-art CNN models such as VGG16, MobileNet, ResNet50, InceptionV3, and DenseNet121. The accuracy achieved by the proposed approach is 16% higher than VGG16, 66% higher than MobileNet, 19% higher than ResNet50, 26% higher than InceptionV3, and, 23% higher than DenseNet121. The proposed deep learning model outperformed the other models when tested on the same dataset and achieves 83%, 33%, and 80% higher accuracy with VGG16. Considering the consistent performance and modular framework of the proposed work, this can be extended to various ITS systems across the world.</p>