• 

  https://lafaekaitidb.vercel.app/index.html

• 

  more about lafaek ai

• 

  Lafaek ai testing demo

• 

  show camera detection map

Inspiration

The inspiration for LAFAEK AI came from the urgent need to improve public safety and law enforcement capabilities in Timor-Leste, particularly in the capital city of Dili. With an increasing number of vehicles and rising incidents of theft, it became clear that traditional methods of tracking stolen motorcycles were inefficient and time-consuming. The idea for LAFAEK AI was born out of a desire to utilize AI and advanced technology to automate the process of analyzing CCTV footage, allowing authorities to respond quickly to criminal activity and improve overall public safety. The project aims to bring cutting-edge solutions to regions that can benefit greatly from technological advancements.

What it does

LAFAEK AI is an innovative application designed to assist law enforcement agencies in Timor-Leste by automating the process of tracking stolen motorcycles. The platform uses AI to analyze CCTV footage, detect motorcycles, recognize license plates, and store relevant data, such as location coordinates and timestamps. By entering the license plate number and date of theft, authorities can quickly determine if the motorcycle has been spotted by any cameras and receive real-time notifications of its location. This significantly reduces the time and effort required to manually review CCTV footage, making it easier for police to recover stolen vehicles and deal with related crimes efficiently.

How we built it

1. CLIP-TiDB Image Embeddings
   We integrated the CLIP model to generate image embeddings, which are then stored in TiDB Serverless with Vector Search. This allows for sophisticated image similarity searches, improving the identification process for vehicles. Read more about this process: LinkedIn Article and explore the repository: GitHub Link.

2. Lafaek AI Landing Page MVP
   We also developed a landing page for the Lafaek AI application, serving as a portal for users to access the app and learn more about its features. Check out the MVP landing page here: GitHub Link and see the live demo here: Vercel Demo.

Challenges we ran into

One of the main challenges we faced in developing LAFAEK AI was the inability to access CCTV data directly from key locations in Timor-Leste's capital city, Dili. As a startup idea, the app was not yet licensed, limiting our access to the city's vast surveillance network. This limitation prevented us from working with real-time data, which is crucial for tracking stolen motorcycles accurately and in a timely manner.

To overcome this hurdle, we resorted to manually recording videos from roadside locations. While this approach allowed us to collect some data, it added significant complexity and limited the scope of our initial testing. The need to manually collect and process the video footage posed additional challenges in terms of time, resources, and data realism.

Despite these difficulties, we remained committed to our vision and used creative problem-solving to simulate real-world conditions as closely as possible. This experience underscores the importance of data accessibility in developing AI-based solutions and highlights the need for strong partnerships with local governments to unlock the full potential of LAFAEK AI. The challenges we have faced have only strengthened our resolve to forge ahead, refine our approach, and ensure that our solutions deliver impactful results, even at this early stage.

Accomplishments that we're proud of

We are very proud of the progress that has been made in developing LAFAEK AI, especially considering the limited resources and most of the work being done independently. Despite being a solo effort, I remained committed to seeing this project through from concept to execution. This journey involved overcoming significant technical and logistical challenges, such as manually collecting data and building a robust AI system capable of analyzing it effectively. The ability to overcome these hurdles and create a functional prototype that has the potential to help law enforcement in Timor-Leste is a testament to dedication, perseverance, and a deep belief in the power of technology to solve real-world problems. This achievement not only demonstrates resilience but also sets a strong foundation for future development and collaboration as the project grows.

What we learned

This project has been an incredible journey, full of significant learning experiences that have brought a concept to life. I discovered the incredible power of TiDB Serverless, which proved to be an ideal solution to overcome local resource limitations. Its robust and scalable architecture allowed us to implement a sophisticated database system without the hassle of lengthy installations. Exploring vector search technology was a first for me, and integrating it into our application provided a deep understanding of its potential in AI-based solutions. Beyond the technical aspects, the hackathon taught me the importance of effectively communicating the progress and value of our project. Writing the LinkedIn article not only helped in sharing our work with a wider audience, but also fostered connections that could be crucial for future collaborations.

What's next for LAFAEK AI

Going forward, the next step for LAFAEK AI is to expand our collaboration and research efforts. We plan to partner with universities in Timor-Leste and internationally to conduct more comprehensive research on this project. Our goal is to produce academic papers or journals that can contribute to the body of knowledge in the field of AI and public safety, with a specific focus on Timor-Leste.

In addition, we aim to obtain permission from the relevant authorities to access real-time CCTV data from Dili. This access will improve the accuracy and effectiveness of our system, allowing us to conduct more precise research and deliver better results. We will also advocate for the adoption of TiDB Serverless, emphasizing its robust security features and cost efficiency, which eliminates the need for expensive infrastructure investments. By showcasing the affordability of TiDB and AWS S3, we hope to convince stakeholders of the long-term financial benefits.

Finally, we envision expanding the application of LAFAEK AI to address broader issues, such as traffic violations, helmet use, and disaster response. Our ultimate goal is to deploy this technology nationwide, helping to prevent crime and encouraging young people to pursue education and meaningful employment.

Instructions for Accessing/Testing Our App

To effectively test LAFAEK AI, please follow these instructions:

1. Date Selection: Select a date range from August 16, 2024, to August 18, 2024. Our data only exists for this specific timeframe, so selecting these dates will provide the most accurate results.

2. Number Plate Number: You can enter any license plate number from 0000 to 9999. For money plates already in the database, we recommend using the license plate number 9824, as this has been previously detected by our system.

3. Brief Description: In the description field, enter "red and black motorcycle ”, or the characteristics of the stolen motorcycle. This will help the system filter and return the most relevant results based on the existing data.

Built With

• amazon-web-services
• apache-nifi
• easy-ocr
• mysql
• openai-clip
• python
• scikit-learn
• streamlit
• tidb
• vercel
• yolov3