Authors: Kawale Pranita Ravindra, Professor Bangar Abhishek Popat, Professor Bhosale Sachin Bajirao
Abstract: Drunk driving is a major cause of road accidents, and it can be prevented. Many low-cost alcohol detection systems use only one breath sensor and one fixed limit. If the reading goes above the limit, the engine is switched off. If not, the car keeps running. This paper looks at what IEEE research papers say about doing this in a better way. We studied 11 IEEE papers (8 journal papers and 3 conference papers). They cover face-based detection, sensor-based detection, engine lock systems that use the MQ-3 sensor, portable breath testers, wearable sensors, and skin conductance sensing while driving. The papers show that deep learning on face images can reach about 87 to 90 percent accuracy, that wearable sensors with machine learning can reach about 90 percent, and that cheap breath sensors can lock an engine and send the location of the vehicle. They also show some problems. Accuracy is lower for some groups of drivers. Even sober people breathe out a small amount of ethanol. Skin signals change with stress and not only with alcohol. And very few papers test what the vehicle should do after alcohol is found. Based on this, we describe GripSense. It is an idea for a system that puts a skin temperature sensor, a skin conductance (GSR) sensor and an MQ-3 breath sensor on the steering wheel and combines their readings. It reacts in two steps: a warning for a low level, and a slow, safe slowdown for a high level. GripSense is not built or tested in this paper. The paper gives a comparison of 11 studies, a proposed design, a list of research gaps and a plan for testing.
