2b || !2b

central server
virtual machine (for running object detection)
Signup/Login
Routing destination
functional app on azure
Map rendering
Location and severity input

Inspiration

The key factor in emergency medical situations is the Initial Response Time. Ambulances are often found meandering their way through highway intersections and gridlocks, losing out on the crucial initial minutes which can prove fatal for the patient. With more than 20% patients losing their lives due to delay on roads, the time that should ideally be spent on providing first response. measures is often whiled away stuck in congestion. Static Traffic Light durations often cause heavy congestion due to inability to adapt to changes in congestion density.

What it does

The proposed idea is that of an ‘Automated Traffic Management System’. The traffic light status is calculated taking into account the following factors :- a. The congestion coefficient is calculated using image recognition applied on live feed captured by CCTV cameras. b. Live location, severity index and bearing angle of the approaching emergency vehicle from the nearest traffic signal. c. Microsoft Azure’s Computer Vision API would be used to check the presence of emergency vehicles in the CCTV live feed. If the presence of any emergency vehicle is detected, maximum priority would be given to change the status of the respective traffic light to green. The traffic light location would be fetched with the help of an API built using government provided database.

How we built it

Remote Server: NodeJs/Flask Client Side Application: Java/Kotlin Deep Learning and Image Processing Model: OpenCV, PyTorch, Fast.ai Real-time Location and Distance: MapMyIndia SDK/APIs Obtaining video frames and image processing: Azure Computer Vision and Video Indexer APIs DL model deployment: Azure VM

Accomplishments that we're proud of

Reducing congestion at traffic signals.

Allowing emergency vehicles to pass without too much disruption in traffic.

Reduce the effective response time of emergency vehicles.

Feeding the collected congestion data to a machine learning model to predict the congestion on a road on any given day. This helps us to identify peak traffic hours of various regions of the city.

Collision resolution in case of more than one emergency vehicle approaching from different directions by using a severity index.

What we learned

Setting up Azure VM and functional apps Using dockerized containers Setting up MapMyIndia SDKs and APIs Multi-process threading Computer Vision API

What's next for 2b || !2b

Feeding the collected congestion data to a machine learning model to predict the congestion on a road on any given day. This helps us to identify peak traffic hours of various regions of the city.

Reinforcement learning for calculating the weights of various factors used for computing durations.

Precise coordinates

Reducing latency by using GPU

Building pipelines from the central server to traffic signal microcontroller to integrate the code to hardware