In this paper, we propose a smartphone self-safety app of two-wheeler riders, that ascertains whether the riders are wearing their helmets properly. Our smartphone app is a self-safety warning system that alerts riders if they are not wearing the helmet or if their helmets are not worn properly. To this end, we have collected a large amount of data under varying daylight conditions and helmet styles. We also propose a novel method for auto-annotation of data to reduce time and effort needed for data labeling. In our work, we have trained deep learning models for the detection of head and helmet for analysing if the rider is wearing the helmet. In case, the riders are wearing the helmet, another deep learning model is trained to classify whether the helmet is worn properly or not. We have also discussed the method to deploy the model on edge devices and effects of model optimization. We have assessed the model performance using multiple metrics to demonstrate the effectiveness of our approach in addressing critical safety concerns for two-wheeler riders.