Inspiration

We look to address the third problem statement – integrating with a smart environment. We will narrow our scope to reporting traffic incidents and we leverage technology to accomplish these three tasks: To make better and more timely sense of emergency traffic accidents. To trigger early intervention measures such as notifying myResponder app users in the vicinity. For SCDF to have a clearer picture of the accident before activating emergency resources such as ambulances.

In 2018, there were 7690 traffic accidents involving injuries. Out of this number, 120 people were killed. Singapore’s road fatality rate of 2.73 per 100,000 citizens is higher than London, Hong Kong and Tokyo. The common causes of such road accidents were due to driving under the influence of alcohol and willfully violating traffic light conditions.

The red bar represents the number of fatalities during accidents, while the black bar represents the number of fatal accidents. [1] Our project solution aims to reduce the probability of fatalities during severe car accidents, and dispatch EMS promptly when major road accidents occur.

According to SCDF’s annual statistic in 2019 [2]:
There are roughly 520 EMS calls per day among 80 ambulances.
~48 EMS calls, or 9%, were false alarms or non-emergencies.
~28 EMS calls were related to road traffic accidents.

It is clear that EMS are stretched thin. Furthermore, we believe that the percentage of non-emergency calls and false alarm calls of 9% is still too high and detracts away from other medical emergencies such as that of road accidents.

Key Issues

  • EMS are stretched thin. We want to avoid activating or flooding their network with calls if the road accident isn’t fatal or serious.
    • Information from road accidents are largely based on the caller. This is problematic for several reasons:
    • Depends on the response time of the caller
    • Caller might be distracted during the call since he/she could be driving
    • Caller may not provide information succinctly
    • Road accidents may occur in poor lighting, or in wet weather, or in a secluded area, where it may be more difficult to report such incidents.

What it does

Solution description

Smart cameras can be used to detect the occurrence of road accidents and automate accident reporting to reduce emergency response time. By leveraging on image recognition technologies, pertinent accident information may be included in the report.

How this works is when the IoT device captures images of the accident, the traffic police will be notified of the accident while SCDF will verify the picture received on the web application. After which, users registered on the myResponder app who are within 1km of the vicinity will be notified of the accident. They will then be able to submit on-the-ground reports to the web application which will serve as a secondary source of information of the accident which will be displayed on the web application. Additionally, medically trained CFRs can also provide early intervention measures during major accidents while the EMS are onroute. Finally, SCDF can broadcast the message to citizens via Telegram and/or deploy EMS if needed. Hence, SCDF and other relevant authorities would be better informed in their allocation of appropriate emergency resources.

This solution serves four purposes:

  • To augment the primary report sent by the server.
  • To provide constant updates of the situation.
  • For the relevant authorities to verify the information sent by the IoT device.
  • For messages to be relayed to drivers on the road.

Justification and direct impact of solution

Currently, information about road accidents are largely based on information from the members of the public who were at the incident site. As such, it may potentially take minutes for vital information to reach the relevant authorities, including the SCDF. Furthermore, since information is heavily reliant on the caller, the absence of witnesses to make the call or the possibility of the caller delivering false information (intentionally or not) would hinder SCDF’s ability to perform at their best. There is, therefore, a need for SCDF to have a new primary means for obtaining more accurate information regarding traffic accidents.

Our solution aims to send information to the authorities as soon as the accident occurs. This reduces the need for them to rely on information given by the public at the incident site. Instead, the reports made by the members of the public will be used to augment the primary report. This solution addresses every aspect of the problem statement. Firstly, since SCDF is notified of accidents within a shorter time frame, they can respond to it sooner than before. Secondly, the messages sent to nearby drivers will alert them of the accident and they can make way for EMS if it is classified as a major accident. This allows EMS to reach the incident site faster and/or potentially helps to ease traffic conditions, serving as an early intervention measure. Lastly, if the accident is scored by the system to be a minor accident and verified by an SCDF employee as such, it would not be necessary to dispatch an ambulance, saving emergency resources. As a result, this allows SCDF and other authorities to respond to the accident in time, potentially saving lives.

How IBM Cloud is used

  • IBM Cloud will be used to train a model to classify accidents, identifying and labelling the images sent by the IoT device accordingly
  • Examples of labels
  • Normal
  • Minor Accidents
  • Major Accidents

How I built it

Solution description

Smart cameras can be used to detect the occurrence of road accidents and automate accident reporting to reduce emergency response time. By leveraging on image recognition technologies, pertinent accident information may be included in the report.

How this works is when the IoT device captures images of the accident, the traffic police will be notified of the accident while SCDF will verify the picture received on the web application. After which, users registered on the myResponder app who are within 1km of the vicinity will be notified of the accident. They will then be able to submit on-the-ground reports to the web application which will serve as a secondary source of information of the accident which will be displayed on the web application. Additionally, medically trained CFRs can also provide early intervention measures during major accidents while the EMS are onroute. Finally, SCDF can broadcast the message to citizens via Telegram and/or deploy EMS if needed. Hence, SCDF and other relevant authorities would be better informed in their allocation of appropriate emergency resources.

This solution serves four purposes:

  1. To augment the primary report sent by the server.
  2. To provide constant updates of the situation.
  3. For the relevant authorities to verify the information sent by the IoT device.
  4. For messages to be relayed to drivers on the road.

Justification and direct impact of solution

Currently, information about road accidents are largely based on information from the members of the public who were at the incident site. As such, it may potentially take minutes for vital information to reach the relevant authorities, including the SCDF. Furthermore, since information is heavily reliant on the caller, the absence of witnesses to make the call or the possibility of the caller delivering false information (intentionally or not) would hinder SCDF’s ability to perform at their best. There is, therefore, a need for SCDF to have a new primary means for obtaining more accurate information regarding traffic accidents.

Our solution aims to send information to the authorities as soon as the accident occurs. This reduces the need for them to rely on information given by the public at the incident site. Instead, the reports made by the members of the public will be used to augment the primary report. This solution addresses every aspect of the problem statement. Firstly, since SCDF is notified of accidents within a shorter time frame, they can respond to it sooner than before. Secondly, the messages sent to nearby drivers will alert them of the accident and they can make way for EMS if it is classified as a major accident. This allows EMS to reach the incident site faster and/or potentially helps to ease traffic conditions, serving as an early intervention measure. Lastly, if the accident is scored by the system to be a minor accident and verified by an SCDF employee as such, it would not be necessary to dispatch an ambulance, saving emergency resources. As a result, this allows SCDF and other authorities to respond to the accident in time, potentially saving lives.

How IBM Cloud is used

IBM Cloud will be used to train a model to classify accidents, identifying and labelling the images sent by the IoT device accordingly
Examples of labels:

  • Normal
  • Minor Accidents
  • Major Accidents

How coding is used

  • The IoT device will transmit information to a server.
  • The server will relay the information to SCDF if it is a severe road accident, and it will also notify users of the application as well as relevant authorities like the traffic police.
  • We intend on leveraging machine learning technique for image recognition, running on cameras, that determines the severity of road accidents to aid the SCDF in the following scenarios:
  • In the event of a non-daylight crash (12:00am – 6:00am) it will form as a passive observer, alerting the SCDF of a crash, when passive observers are scarce.
  • Includes priority levels of the crashes, to allow EMS to tend to the most severe crash when/if their resources are already stretched.

Challenges I ran into

Possible Challenges and Mitigation Strategies

Ethics behind usage of machine learning in the classification of road accidents

  • Might be unethical to show faces of the victims in a road accident. A mitigation strategy would be for the server to blur the faces of the victims once results from the model is received.
  • The model may not be able to classify the severity of road accidents with absolute certainty, resulting in false negatives and consequent less-than-appropriate emergency medical response. It is unethical to trust and allow AI to classify such accidents as these are life-and-death situations, and should require further human action to verify and evaluate an appropriate response.

IoT device may not have sufficient bandwidth to send images to server

  • LTE Cat.M1 has a relatively higher maximum transmission rate that is appropriate for photo transmission. [4]

Solution does not yet take into account the possibility of multiple accidents occurring at the same time.

  • In addition to the current labels on the data, we can include the severity level of the accident. The relevant authorities can then dispatch resources accordingly based on the severity level as well as the location of the incident.

One-way reporting limitations

  • A mitigation strategy would be to allow for two-way communication between responders on the application and the relevant authorities, possibly via the use of a chatbot that keeps track of the current situation.

Accomplishments that I'm proud of

Final Summary of how the solution directly responds to the problem statement

The solution leverages on IoT devices and machine learning to:

  • Classify road accidents when it happens.
  • Send reports to relevant authorities and to the application.
  • Continually provide updates from the members of the public at the scene.

This solution reduces the time taken for vital information to reach authorities like SCDF, allowing them to quickly trigger early intervention measures if needed and activate emergency medical resources if necessary.

What's next for SCDFxIBM Lifesavers Innovation Challenge: Call for Code 2020

Future Improvements

The IoT device can include other labels:

  • Traffic conditions
  • Number of casualties
  • Number of cars involved in the accident
  • Weather condition This also requires the IoT device to include other sensors to detect weather conditions like rain.

Built With

Share this project:

Updates