Accicare
Accicare is the ideal solution for reducing accident-related deaths from happening on roads. Accicare is for every car, so the next drive doesn't have to be with the fear of death!
Injuries and deaths resulting from road accidents are a growing public health problem in India. Road crash deaths have increased by 31% from 2007 to 2017. Surveys show that a 10 min reduction of the medical response time can be statistically associated with an average decrease in the probability of death by one-third, both on motorways and conventional roads. We propose an accident mitigation system that can be used in post-accident vehicles to reduce the death rate due to road accidents.
Accidents are imminent in vehicles. Therefore, stopping them and preventing them is a task that cannot achieve fullness. The alternate approach is to prevent death due to accidents rather than preventing the accident itself. To reduce the death rate due to road accidents, we propose a system that will make a hybrid decision based on both the Machine-learning model and human input to identify if the victim of the accident requires medical help or if they are safe. The medical help request decision made by the hybrid decision approach checks for 2 conditions to finalize the decision.
1. Presence of blood
2. Consciousness
Once these conditions are checked using the input from the sensors, the system uses information from the user to decide if the passenger/driver needs medical help or not.
The solution doesn’t end with just identification of the victim’s need for medical help but includes intimation of hospitals to take quick action for the emergency. The range identified for quick emergency response from hospitals is 5km. The 5km range enables ambulances to arrive at the accident location well within 10 minutes of the accident making the solution adequate and highly responsive.
Presence of Blood
The video stream from the mounted camera is cropped to required size. Once the images are cropped to the required size, they are passed to a BloodFilter that extracts the blood feature from the image. By doing this I found that the Deep Learning CNN model gives predictions with higher accuracy. MobileNetV2 architecture is a conventional neural network which is a Keras image classification model, loaded with weights pre-trained on ImageNet as a reference for Transfer Learning. Since the model is used for binary classification, binary cross-entropy is chosen as the loss function. Additional layers with ‘Relu’ and ‘SoftMax’ as activation functions are used to improve the accuracy further.
Consciousness
The second condition the system checks for to confirm its decision is consciousness. The consciousness of the victim is decided based on both visual data and voice responses from the user. These inputs are converted into a consciousness score that can have a maximum value of 6 and a minimum value of 2. A score of 6 means that the victim is fully conscious while a score of 2 means that the victim is unconscious.
Hospital Finder Program
The program is deployed in the cloud, The program aims to find nearby hospitals in the specified 5km range around the accident location coordinate. Here, I use Google Maps API. Using this API, we find nearby hospitals based on the coordinates.
Portal
The portal is designed to send information on the accident to hospitals as managed service. Every hospital is registered in the portal with login credentials to ensure the safety of accident information. The portal gives hospitals all the vital information such as the need for medical help, consciousness score and the victim’s response to predefined questions that indicate further clarity on the victim’s state. The hospitals are to respond to the information in the portal such that there are no resources wasted on the same accident. The portal gets updated in real-time and will indicate other hospitals in the same range when a hospital responds to the accident.
LIVE WEBSITE: https://accicare.github.io/ACCICARE/index/
Project Numbers
2
Mentors
5
Members