Bike riders are increasing in number among other vehicles. The current traffic routine survey shows many of accidents are being caused by bike riders and many of them are dead. These factors have driven recent research in applying intelligent and computerized approaches to detect bike riders with helmet so lives of bike riders can be saved. However, these systems are not viable as they require significant time or compute power to process high-resolution images. This research objective was to investigate bike rider for helmet detection via deep learning techniques using different images from different size, angle, and scene. The intent was to reveal whether machine learning models could be developed that provide high confidence results using fractional resources by using images. A deep learning pipeline has been developed which reduces high-resolution to a sufficiently small size so they can be fed as input into a CNN for binary classification (i.e., bike rider with helmet or without). Several improvements have been implemented to boost general performance, namely supplementing the training data, and adding data augmentations. Ultimately, the developed low-resolution model is effectively skill-less for very low-resolution inputs. An observed significant decrease in model inference time is a superficial benefit given the loss in classification capability. This is a promising and expected result arising from an obvious limitation with the methodology. In conclusion, the developed deep learning pipeline is suitable as a viable helmet detection system when using the input resolutions investigated. Proposed method has performed best among state of the art methods and provide 97.34% of accuracy.