<p>Motorcycle usage is growing globally and one of the main reasons is affordability. Despite helmet laws, many riders don’t wear them. Not wearing helmets may enhance the threat of brutal head trauma, death and coma. Researchers have applied surveillance video systems to detect traffic violations, particularly for riders without helmets. While deep learning algorithms for helmet detection in road surveillance have been developed, less focus has been given on adapting them for the complex environments of construction sites. Most of the traditional methods of helmet detection rely on image processing or manual inspections, which can be time-consuming and inaccurate. Hence, an automatic motorcycle and helmet detection mechanism is implemented in this paper with the help of deep learning strategies. At first, the requisite videos are fetched from the available data sources. From the fetched videos, the motorcycle is detected with support of Adaptive Yolov5 network. Here, the network performance is improved by optimizing the parameters using Improved Hermit Crab Optimizer (IHCO). Subsequently, the Region of Interest (ROI) is extracted from the detected video frames. Furthermore, the Vision Transformer-based Dilated Residual DenseNet (ViT-DRDNet) is introduced to recognize the helmet from the ROI extracted frames. If there is no helmet detected in the specific motorcycle, then the number plate of the motorcycle is recognized by Adaptive Yolov5 network. Based on the detected number plate of the vehicle, the message alert is sent to the registered phone number, which facilitates the secured driving system. Lastly, the performance and supremacy of developed motorcycle and helmet detection mechanism are estimated among various existing techniques and algorithms. The accuracy score of the proposed helmet detection mechanism is 94.95, showing its overall ability in enhancing the road safety of the public with accurate detection outcomes.</p>