Best Main Prize - Machine Learning - Vigil

Life saving application capable of recording and providing immediate help for 13 different anomalies ranging from accident, theft, assault to arson,burglary and kidnapping.

Comment

Inspiration

Currently videos are recorded through CCTV cameras, and manual surveillance is undertaken which requires a lot of human intervention and there is no facility for real-time emergency support. The services provided are excruciatingly slow which can have very bad result in cases of accident or fire etc. leading to several deaths around 100,000 annually and excessive loss of property.

What it does

  1. Speed up the deliverance of services
  2. Use an Anomaly Detection Deep Learning model
  3. Detect events such as fire, robbery, accident.
  4. A mobile application for the citizens
  5. A web application to monitor the intensity and service request.
  6. Crowd–Source to reduce false alarms In a nutshell, saves life in situations where every second counts. ## How I built it Proposal to speed up the deliverance of services and help like, Police Backup in case of theft and robbery, hospital aid in case of accidents etc. via IoT coupled with Fog and Cloud Architectures, and effective surveillance. Various unusual events are detected (with the help of Deep Learning techniques) sourced from single static cameras and notifying the concerned authorities about the event with its timestamp and additional details like extent of damage, parties involved and location which assist the authorities in identifying the cause of the anomaly and take necessary action. The anomalies proposed to be detected include: • Abrupt Fire • Accident • Burglary • Physical Assault • Robbery • Shop Lifting • Traffic Rules Violation In case of events requiring an immediate attention, we propose to notify the crowd in the vicinity of the region where the event has occurred so that they can voluntarily choose to help which can minimize latency in saving a life or even catching a perpetrator.

Challenges I ran into

  1. Acquiring Dataset
  2. Heavy Computational power ## Accomplishments that I'm proud of Able to achieve a low false rate of 1.9, which allows classifying most of the anomalies correctly.

What's next for Vigil

  1. Integration Fog Computing for decreasing latency.
  2. Deploying on real ground.
Share this project:

Updates