Ridehailing companies suffer from impersonation and we found out that there had been a potential that one of the leading Ridehailing companies could be blocked in London for drivers' identity theft and hazardous behaviour.
What it does
It detects the various violations by drivers (impersonation, traffic rules violation, health and security hazards) and alerts the control centre staff for them to take necessary actions.
How we built it
We propose an integration to the existing Swvl application which helps identify bad actors in real time using Neural Networks and conventional Computer Vision algorithms.
Moreover, we also crowdsource the customers’ feedback on our alerts to build a sense of trust and ensemble our performance metrics by increasing the severity levels of alerts and prioritising the alerts to the control centre for taking necessary actions.
We used various ML models to detect the various violations by drivers and used mobile, backend and frontend technologies to develop an end-to-end working flow.
Challenges we ran into
Streaming the real-time video to the backend from driver's side. Getting good results from ML models for different use cases.
Accomplishments that we're proud of
We managed to develop and end-to-end working PoC in just ~2 days with the close collaboration among us.
What's next for Safeti
- We can extend our face detection and recognition work to simplify the customers’ onboarding on to the vehicle.
- We can send driver’s GPS coordinates to approximate the current velocity and check for over/under speeding.
- We can quantize our ML models to run them on driver’s device, and save data/compute.