s per a report published by the World Health Organization, 94% of the fatal traffic accidents, amounting to 1.25 million deaths, are caused by driver error. By 2030, road traffic injuries are projected to be the fifth leading cause of death worldwide , surpassing HIV/AIDS, all forms of cancer, violence, and diabetes. As per the study, drivers who send and receive text messages, take their eyes off the road for an average of 4.6 seconds out of every 6 seconds while texting. At 55 miles per hour, this means that the driver is traveling the length of a football field, including the end zones, without looking at the road.
These distractions are mainly due to Visual, Manual or Cognitive Tasks each of which can be avoided, if there is a proper training and surveillance involved. To solve this, a leading garbage truck manufacturer in the US was spending heavy on driver training and identifying a driver's distraction using in-cab recorded videos. In addition, it had to employee 100+ part time employees to watch these videos and tag them to the predefined driver distraction categories that would help in the driver trainings. This is expensive and yet less effective.
We designed an automated process to monitor the driver in real-time using Deep Learning and Computer Vision and enabled in-cab alert management.
The in-cab camera captures real-time feed of the driver which is pre-processed on a single board computer using computer vision. The processed image is then classified by a neural network model, trained on a few hundred images per scenario, into one of the 5 types of distractions viz. smoking, drinking, texting, blocking camera, without seatbelt. Once the neural network model identifies a distraction, the driver is immediately alerted preventing any accidents and at the same time an event is generated to the hauler and archived for driver’s training.