AI Camera-based driver behaviour analysis

  • project: Driver's behaviour analysis based on AI
  • client: Sparkbit internal project
  • year: 2019
  • website: www.sparkt.ai

challenge:

The goal of the project was to create a next-generation driver behavior analysis platform for motor insurance purposes. Unlike all other existing platforms, the solution should utilize truly contextual data, coming from a camera.

Additionally, the solution should be cheap to introduce to B2C customers, thus it cannot require additional hardware

tools used:

Tensorflow,
Keras,
YOLO,
CNN

We wanted to create the most advanced on the market driver's behavior analysis based on camera image processing.

solution:

We have created a module that uses images from a smartphone camera to analyze dangerous road situations, such as tailgating, hopping between lanes of traffic or speeding in the proximity of pedestrian crossing.

We have built a deep learning model, using the CNN model and YOLO detector.

We have used transfer learning from a model trained on COCO and then trained it on a custom-built dataset (ca 50 000 images captured from a dash camera) on our specific use-case.

impact:

We created a system which with the use of images from a smartphone camera can analyze dangerous road situations and detects events such as tailgating, hopping between lanes of traffic or speeding in the proximity of pedestrian crossing.