<p>In today’s times, the escalating rate of powered two-wheeler (PTW) related accidents and fatalities present a significant global road safety challenge. Current Advanced Rider Assistance Systems (ARAS) that are commercially available, are expensive and rely on specialized hardware. This has restricted their adoption to the premium motorcycle market. In this paper, we propose a lightweight and low-cost novel ARAS on a smartphone for the safety of PTW riders that works in real-time. Our ARAS is a self-safety alert system that leverages on the multiple sensors available on a smartphone to provide two critical safety features: pre-ride helmet detection and proper wearing of the helmet, in case detected and while the PTW is in motion, it performs real-time stability recognition and fall detection. It alerts the rider whenever it finds that the PTW is unstable and if a fall occurs, it communicates for help. To this end, we have prepared two novel video datasets for helmet detection and proper helmet wearing classification and fine-tune a deep learning model for these two tasks on our data. For stability recognition, we use publicly available multiple sensor data and extract specific statistical features as well as temporal features from the LSTM model and combine these for robust stability recognition and fall detection from time-series data. We have assessed the model performance using multiple metrics and also compared our results with state-of-the-art methods to demonstrate that our system provides a crucial, missing technological layer that complements and strengthens the existing safety initiatives for PTW riders, especially, in low-to middle-income countries (LMICs).</p>