rideCare

Apprehensive of your safety in a taxi? Worry no more!!! If you rideshare, rideshare with rideCare!

Comment

Inspiration

The growth of ride-hailing services has revolutionized the transportation industry, but it has also raised concerns about the safety of riders. In order to address these concerns and provide a secure and enjoyable experience for all riders, we came up with an innovative solution to monitor the rides using a hardware device that automatically triggers a dashcam to stream the happenings live to authorized personnel, and sends a text using Twilio to an emergency contact person of all the people in the car, so that they can use manual intervention and cognizance to cater to the safety of the people.

What it does

rideCare is a system that integrates Machine Learning Techniques, more specifically Natural Language Processing with a hardware device such as the Raspberry Pi. It detects conflicts in conversations between the driver, and passengers to broadcast an alert to their registered emergency contacts with a link to view and read the transcript of the conversation, in real-time. It can also be used by 911 to intervene as they seem fit.

How we built it

Hardware Used:

  1. Raspberry Pi Model 4 B (Running Raspbian OS)
  2. C922 PRO HD STREAM WEBCAM
  3. Blue Yeti USB Microphone #Software and Packages Used:
  4. Python 3
  5. Vosk for Speech to Text Conversion
  6. Perspective API for toxic comment classification
  7. OpenCV for streaming the video in real-time
  8. Twilio for sending SMS to emergency contacts

Challenges we ran into

  1. Merging the whole project, especially integration of React with Flask
  2. Use of a textual comment classification library such as PerspectiveAPI to get better predictions when the real-time input is in speech
  3. Sending an SMS to the emergency contact by using Twilio, when the trail account had only toll-free number attached to it which logged the error as 'Undelivered'
  4. Designing a formula (out of 5) to calculate the severity of toxicity in the speech, based on the publicly available Kaggle dataset used by PerspectiveAPI - Kaggle dataset by weighting the various labels to determine the threat of the conversation

Accomplishments that we're proud of

  1. Use of OpenAI API to also perform sentiment analysis, in addition to PerspectiveAPI
  2. Use of a speech-to-text library such as vosk to convert speech to text in real-time to make predictions using text as input

What's next for SafeRide.ai

  1. Use of speaker diarization to identify targeted emergency contacts
  2. Development of a full fledged web-page to stream data

Built With

Share this project:

Updates