This paper introduces an enhanced deep learning model tailored for the precise detection of motorcycles in challenging urban scenarios, particularly where occlusion frequently occurs. It includes those with significant levels of occlusion. Our Proposed Work Region-based Faster R-CNN architecture is demonstrated to work efficiently in visually obscured environments. Extensive training of the model was performed on two different datasets: the 10,000-annotated Urban Motorbike Dataset (UMD-10 K) and the newly created Motorbike Dataset (SMMD), which contains 5,000 images obtained from the traffic Control CCTV system. Our method exhibited excellent performance on the UMD-10 K dataset with an average precision (AP) of 88.8% even though 60% of the motorcycles are partly occluded, and low-angle, mobile cameras are used for recording. In contrast, for the SMMD dataset, we got an AP of 79.5%. We show that Proposed Work outperforms existing GOA models, such as YOLO V3 and a Faster R-CNN model VGG16 based, which demonstrates the potential of Proposed Work for application in real-time monitoring systems for urban traffic behaviors.