Sandhya Gopalan

Sandhya Gopalan

Sandhya Gopalan

Preventing accidents through real-time warning of dangerous Spots to connected vehicles

Every year 20-50 million people face road accidents and over 1.3 million people lose their lives from road accidents throughout the world. Majority of the road accidents are caused by distracted driving, drunk driving, reckless driving, rain, running red lights, unsafe lane changes and poor road quality conditions. The current approach of identifying accident prone zones are based on traffic and accident statistics that are outcome of driving behavior and road quality conditions. This approach works on hindsight and depends on the aftermath of accidents. To overcome this, we have come up with a new approach enriched on the past and current driving behaviors and road quality conditions to avert accidents through real time warning by leveraging connected smart vehicles. Our approach to prevent accidents through real time warning is by labeling various road spots based on their degree of accident proneness. The accident proneness of a road spot is determined by the concentration of reckless & aggressive drivers passing by and quality of the road. In our experiment, 100 cars were embedded with sensors that were transmitting data to a centralized server for 6 months. On every GPS location of the road, the behavior of the driver was assessed based on speed, adherence to traffic rules, braking style, wheel maneuver under different weather conditions under different time periods. With the former derived behaviors as features, supervised machine learning and unsupervised clustering algorithm were used to identify the Aggressive and Reckless drivers. The road quality index was derived from accelerometer data of connected cars using random forest and boosting methods. The blind curve faced by drivers in various spots, intensity of traffic violation in major and minor road junctions were identified using Bayesian machine learning techniques from the driving data. By combining the driving behavior and road quality features derived through analytics on data from connected cars, every spot on a road was labeled as either safe, risky or hazardous.