Home / Projects / HacX! Hack for Public Safety — VISTA
← All projects

HacX! Hack for Public Safety — VISTA

2024 Hackathon
YOLOv8AzureComputer VisionReinforcement Learning

Top 10 Finalist — Team Triple L

Issued by: Home Team Science and Technology Agency (HTX) and Microsoft

Top 10 Finalist out of 100+ teams. We developed VISTA (Vision Integrated System for Tool Accountability), a technology-driven solution for Prison Services to streamline tool management using computer vision and reinforcement learning, estimated to reduce manual processing time by 75%.

Team at HacX

The Problem

In prison environments, every tool brought in by contractors must be meticulously tracked for security. The existing manual process of logging, verifying, and assessing tools is time-consuming and prone to human error.

Introducing VISTA

VISTA consists of two integrated components: a Tool Detection Unit and an Online Form, working together to automate the entire tool accountability workflow.

Introducing VISTA

VISTA Tool Detection Unit

A physical detection station equipped with:

  • High-Definition Camera — for computer vision implementation using YOLOv8 for real-time tool recognition
  • Weight-Sensitive Platform — for accurate detection of tools placed on the unit
  • Display Panel — for clear and concise instructions to users

VISTA Tool Detection Unit

VISTA Online Form

Once a tool is detected, the form is auto-filled by VISTA — contractors and officers only need to carry out verification. This increases efficiency and reduces human error in the logging process.

VISTA Online Form

Risk Assessment Score

A risk assessment matrix combining “Severity” and “Likelihood” assessments to categorise different contractor visits. Form data is utilised to generate a precise risk score for each contractor visit, enabling:

  • More targeted security measures
  • Informed decision-making based on tool count, location, and recognition confidence

Risk Assessment Score

Technology Stack

Microsoft Azure, Copilot, YOLOv8 computer vision model, and reinforcement learning algorithms.