How can wearables be used to avoid traffic accidents using technologies like AI/ML/IoT ?

4 min readJun 1, 2022

Photo by Alexander Schimmeck on Unsplash

I got the opportunity to participate in analytics competition conducted by Accenture Digital during my first year MBA at IIT Kharagpur. This article is based on our submission. I would like to thanks my teammates Shubham and Abhishek for their contribution.

Before we can suggest ways to prevent traffic accidents, we should find the underlying reasons which are responsible for road accidents. Following are the major reasons for road accidents apart from vehicle failure:

Over Speeding (~65%)
Drunken Driving ( ~3% )
Distractions to Driver (~3%)
Aggressive Driving (~10–15%)
Drowsiness (~10–20%)

Now we will see how we can apply technologies to tackle prevent accidents.

Over Speeding

The whole road network of the country can be classified in different threat zones depending upon the no. of accidents, topography, demographic, vicinity to healthcare, and population density. For the geographies for which clear data is unavailable, ML techniques can be used to assign a threat level (target) to that particular geography based upon the available information on no. of accidents, topography, angle of turn, frequency of lane change and demographic (feature set) . This data then can be integrated with Google Maps, segregating different geographies into different threat levels w.r.t. the predicted threat level.

Smartwatch/Fitness Band with GPS can be used to detect accident prone geographies having higher threat level by communicating with Google maps and then it can be connected with vehicle using IOT which will limit the vehicle speed in those zone (putting a cap on the upper limit)

Drunken Driving

Alcohol can cause a temporary increase in heart rate and blood pressure. All smartwatch are equipped with heart rate sensor. These heart rate data can be analyzed in real time and machine learning model can be used to predict whether person is drunk or not.

Alcohol consumption relaxes muscles all over the body, causing the pupils to dilate as the muscles in the iris expand. This could be detected via smart eye-wear and the data can be coupled with the data from smart watch to detect drunk driving. The combination of data from both smart watch and smart eye-wear will give less false positives and be highly accurate

Distractions to Driver

The GPS data collected by the smart watch coupled with the radial motion detected by watch’s gyro sensors could predict whether the user is driving or not. This information can be used by the watch to temporary block phone calls and other notifications, preventing the user from getting distracted

Aggressive Driving

The GPS data from the smart watch and rotation of the hand along the horizontal axis will confirm the wearer is driving and give an indication of the vehicle’s speed. This real time collection of data can be coupled with the IOT sensors in the car detecting frequent sways from the vertical axis of the vehicle [4]. In majority of cases high speed coupled with frequent sways and braking is aggressive driving. A classification algorithm trained with the above real-time data will be able to detect cases of rash driving with high accuracy. Based on above a warning alert can be generated in the wearable and the vehicle’s IoT system. To improve the accuracy of the prediction a more comprehensive feature set of variation in heart rate and body temperature can also be added to the parameters of speed and gyroscopic sways. Functionality can be added which will inform parents/police in-case of rash driving.

Drowsiness

Design of the proposed real-time wearable driver drowsiness detection system is shown in Fig.1. In the system, the driver drowsiness detection is mainly determined by the driver’s behavior that is extracted from the motion data collected from the smartwatch built-in motion sensors. First, the users define which hand they are wearing the watch on; based on this choice, the relevant SVM (Support Vector Machine) model is loaded, thus accounting for the different features during classification. After the initialization, the motion data (the linear acceleration and the radial velocity) are collected and the features of interest are extracted based on the user-selected definitions; these features serve as an input to the SVM classifier, as mentioned above. If the user status is classified as drowsy, an alarm will be triggered in the form of a vibration. [5]

Thanks for reading. You can connect with me on LinkedIn.