Efficient training of object detection models in road scenes is extremely necessary in order to build some productive autonomous as well as driving assistance systems. But detection of moving vehicles in adverse weather conditions such as rain, fog, snow, and dust is quite challenging as these weather conditions hamper the quality of images. Therefore, the pre-existing state-of-the-art object detectors trained on clean and good weather images give a poor performance on in this scenario. This research work dives primarily into adverse weather object detection using CNN models. Deep learning based YOLO (You Only Look Once) model uses K-means clustering to estimate bounding boxes. Amongst various open source datasets readily available for training and validating even the customized model, DAWN (Detection in Adverse Weather Nature) dataset has been used in which over 1000 real-world bad weather traffic images have been captured. The data has been distinguished into four main weather categories—Rain, Sand, Fog, and snow. Images have been annotated for 6 classes which are significantly lower than default classes available in COCO dataset, i.e., 80 classes, which significantly reduces the respective model train time and scope of classes. Models are trained on pre trained-weights of Darknet-53 Convolutional Neural Network to speed up the training process. Finally, the model has also been assessed and tested on the images downloaded from open source websites.