Objectives

The company transports its bulk and packaged products through commercial vehicles. There are a few set procedures to select transporters, select and train drivers, conduct route risk assessments at periodic intervals, install OBC and fatigue and distraction devices to check driver behavior, coach the drivers, etc.

Presently, AI-based cameras are installed on all commercial product vehicles that only monitor if the driver is exhausted or distracted while driving. The objective is to analyze the videos and perform secondary classifications like microsleep, drowsiness, absent-minded driving, talking on mobile, smoking, looking out the window for a long time, etc., automatically by the AI/Deep learning algorithms. Also, the video resides on a 3rd party cloud server.

Challenges

• Classify within 2 to 3 minutes from the time the video is received.
  • Currently, there are 110-120 K videos received per month.
  • The application must handle up to 10x the volume compared to the current volume of videos.
• Process multiple videos in parallel to reduce the response time.
• The classification accuracy should be greater than 85%.
• The system should work automatically without any manual intervention.
• Threshold parameters/settings should be configurable and not hardcoded.

Solutions

• We built a scalable, robust AI-based driver’s activity recognition system with high accuracy.
• Pre-process the frames before passing them into the AI model.
• A trained AI model classifies video clips received from the current camera system.
• An integrated solution, very well fitted with the existing environment of people and systems.
• An automated SMS or email triggers the respective supervisors for alerts.