TheftWatch

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

Inspiration

One of our team members has firsthand experience working in retail, where shoplifting was always a looming threat. We wanted to help small business owners take back control of their stores and feel more secure. TheftWatch aims to reduce the anxiety and losses that come from theft by providing better insights and actionable alerts.

• 🚨According to the National Retail Federation (NRF), shoplifting cost retailers $112.1 billion in 2022 alone in the United States.]
• 🚨Organized crime groups have become increasingly involved in shoplifting, targeting high-value items and utilizing sophisticated techniques to evade detection.
• 🚨 Studies indicate that shoplifting is a common occurrence, with some estimates suggesting that it affects a large percentage of retail establishments.

Smart cities often incorporate technologies such as:

• 📸 Surveillance Cameras: These can be used to monitor store entrances and exits, identify potential shoplifters, and provide evidence for law enforcement.
• 📊 Analytics and Data: Advanced analytics can analyze patterns in shoplifting behavior, identify hotspots, and predict potential incidents.

How can smart city technologies be effectively integrated into retail environments to deter shoplifting?

What it does

We decided to go much more further and make our own live representation of a store.

1. Used our workspace to simulate a store environment

We even tested on the MLH Stand!

1. Position 4 cameras (our mobile devices) on each corner to simulate surveillance cameras of a store

2. We recorded 5 samples from each camera view to get insights from our ML model

3. Created different graphs that users can view to see shoplifts over time

4. Created a heatmap to see the most dangerous areas of the store to have a clear view of dangerous

After running this experiment on our workspace we realized that the information is very valuable for security staff from the event

• ✅ Upload security footage videos to analyze and provide insigths to store owners
• ✅ Stablish LLM conversations with the security footage to learn more about the shoplift
• ✅ 2D mapping and heatmap of the store to view dangerous zones
• ✅ Facial recognition and timestamp snapshot when dangerous activty is being detected
• ✅ Live dashboard to view data fetched from security cameras
• ✅ Created and trained our own ML model from open data to improve detection: dangerous, suspicious, safe

How we built it

We leveraged computer vision tools and machine learning models to classify customer behavior and identify suspicious actions. The system integrates with WhatsApp for real-time alerts and uses object recognition and facial recognition technologies to provide clear insights into incidents. The intricate dashboard was created to give shop owners actionable data in a simple, visual way.

Challenges we ran into

Developing an accurate classification system for customer behavior was challenging. Defining clear metrics for what constitutes suspicious or dangerous activity, while minimizing false positives, required extensive tuning and testing. We also faced difficulties integrating multiple technologies into a seamless product.

Accomplishments that we're proud of

• 🎯 Hosted and live website
• 🎯 Visual representation of our workspace as a store
• 🎯 Creation of our own ML model to detect shoplifting
• 🎯 Live alerts whenever our cameras detect suspicious activity

We're proud to have built a functional system that addresses a real problem for small businesses. Creating a reliable classification model for customer behavior and successfully integrating different tech components—including real-time notifications and a detailed dashboard—was a huge achievement for our team.

What we learned

We learned a lot about the complexities of building a computer vision solution that works in real-time and can make meaningful decisions based on nuanced data. We also gained insights into balancing the need for accurate detection with ensuring a positive customer experience.

What's next for TheftWatch

We're planning to refine our classification model to improve accuracy and reduce false alarms. We also want to add more features to the dashboard, such as predictive analytics to help shop owners anticipate theft risks before they happen. Additionally, expanding our notification system to integrate with other platforms, such as SMS or email, is on our roadmap.

SDGs 11 & 9: Sustainable Cities, Communities, and Industry Innovation

TheftWatch contributes to both SDG 11 and SDG 9 by promoting safer and more resilient urban environments while fostering innovation and digital transformation in the retail industry. By helping small businesses reduce losses from theft, TheftWatch strengthens the economic stability and safety of cities, enabling more sustainable and inclusive communities. It also empowers smaller enterprises by offering affordable, tech-driven solutions that enhance security and efficiency, directly supporting the development of resilient infrastructure and fostering innovation in retail. This solution provides small businesses with the tools they need to proactively manage risks and optimize their operations, leading to a thriving, secure urban ecosystem.

Smart Cities

TheftWatch aligns with the vision of smart cities by contributing to the digital infrastructure that enhances urban living. In smart cities, data-driven solutions play a significant role in improving safety and security, and TheftWatch integrates seamlessly into this ecosystem. By providing real-time theft alerts and insights, your project helps create smarter, more secure retail environments, contributing to the overall intelligence and efficiency of city infrastructure. It supports the development of smarter business operations while ensuring safety and resilience in urban areas.

Built With

• cloudflare
• mongodb
• openai
• opencv
• roboflow
• streamlit
• tensorflow
• tillio
• yolov8