AI to reduce Car Accidents

AI-Powered Camera Device to reduce accidents for a car rental company.

Problem Statement

The Car-rental company provides self-drive car rental services in India. These cars can be rented by anyone above the age of 18 with a driving license.

There are couple of reasons for the accident rate for self-drive car-rental companies to be more than double of personally-owned cars.

Firstly, car ownership is still very low in India compared to US and European nations, implying a good chunk of people driving these cars are amateur drivers. Secondly, people tend to drive a rented car much more rashly as compared to their personal car. Thirdly, driving in India is not organised and disciplined (compared to Western countries), causing it to be more accident prone.

The Product

The device comes with 4-HD cameras with Infrared technology. The device can be mounted above the rear-view mirror with 3M tape, it takes power from the car battery. The device has 40 hours of video storage and 4G connectivity to stream videos to the cloud. The device also comes with Accelerometer and GPS sensor.

The Tech behind the Product

The device is powered by AI-powered image processing algorithm which helps it to detect cars, bikes, and pedestrians. It calculates the relative speed between objects in front and the vehicle its installed in through a combination of image processing algorithm and GPS. Through relative speed it calculates estimated time for collision and alerts the user through an high-alert beep in case of likely situation.

The device also detects events amounting to rash driving, viz., harsh acceleration, harsh braking, rapid cornering, and overspeeding. The device can send across an audio warning to the user to warn against risky driving behavior.


Reducing rash driving: We want to reduce rash driving behavior by giving real-time audio alerts to the driver. We started the experiment to test out the efficacy of real-time audio alerts by running an A/B experiment with 4:1 control/test group. I would be updating the experiment results here in couple of months.

Proactive action for Accidents: We get real-time alerts in cases of major accidents (the device measures G-force). These accidents are forwarded to the Accident & Breakdown team. These alerts are sent along with an accident report, this report contains the video, GPS location, and the details of accident. This report is acted upon to reduce the response time for Road-side assistance.

Impact Metrics Monitored

Reducing rash driving: Reduction in Accidents, Reduction in number of Alerts, and Reduction in duration of Alert.

Proactive action for Accidents: Reduction in Response time for the Accident.

PS: I am currently a Product Manager working in IoT team in Zoomcar. Zoomcar leads the self-drive car rental market in India. We have partnered with Netradyne to install DriverI devices in the car. This is 1-year long study and I would be updating the article with insights as they come.