About the Client

client logo

The client is a U.S.-based technology provider supporting law enforcement with surveillance and license plate recognition solutions. Its system enables patrol units, detectives, and command staff to identify and prevent crime, allowing officers to scan over 30,000 plates monthly using mobile readers for efficient, contactless compliance.

Country

USA

Industry

Artificial Intelligence

Business Situation & Requirements

The client was working with existing ALPR systems that involved high operational costs and had limitations in handling high-speed vehicles. There were also considerations around data consistency and hardware compatibility, which influenced performance across certain law enforcement scenarios.

The client aimed to build an ML-driven ALPR application to improve speed and accuracy. While exploring experienced AI partners, they discovered Daffodil Software’s AI Center of Excellence (CoE). They required real-time plate detection, instant vehicle insights, continuous model improvement, strong on-device processing, and the ability to handle image inconsistencies while adapting to evolving hardware and data needs.

Daffodil was selected as the technology partner and created a clear development roadmap based on the client’s requirements:

List icon

Develop trained models for instant license plate scanning within milliseconds

List icon

Retrieve the state of origin, vehicle make, and model instantly

List icon

Train and retrain models for progressively faster predictions

List icon

Enable most data processing on the client-side

List icon

Calibrate for image issues like noise and improper orientation

List icon

Update models regularly to maintain compatibility with hardware upgrades

Solution

The Daffodil team developed an ML-based web and mobile application designed to scan license plates of moving vehicles and extract relevant vehicle and ownership details. The solution was built to support law enforcement personnel with timely and accurate insights using existing mobile devices and surveillance systems.

The application focused on enabling real-time detection, faster data access, and improved decision-making in the field. By combining custom ML models with efficient processing, the system was designed to deliver consistent performance across varied conditions while supporting ongoing model improvements and operational efficiency.

The ML-powered mobile and web application was built with the following key capabilities:

icon Custom ML Model

  • Built using TensorFlow with custom training loops to create and train models for accurate license plate detection and recognition tasks
  • Detects vehicles, localizes license plates, and recognizes characters through a structured ALPR pipeline using deep learning models
  • Uses eighty percent of the dataset for training and twenty percent for validation to ensure balanced model performance and reliability
  • Splits the dataset into structured records using defined instructions to support efficient training, validation, and testing processes

icon Instantaneous Data Retrieval

  • Captures, compares, and analyzes license plates in real time to identify vehicles and trigger alerts for law enforcement agencies
  • Notifies the nearest law enforcement agency or department within seconds when a flagged or wanted vehicle is detected
  • Verifies state, case relevance, and associated details to ensure the accuracy and validity of detected license plate information
  • Enables search based on vehicle make, model, and color, with instant retrieval of relevant and actionable information

Instantaneous Data Retrieval
icon Cost-Effective ML Validation

  • Uses TensorFlow to perform client-side validation, enabling the application to run machine learning models directly on user devices
  • Processes live video feed and performs real-time inference without relying on external APIs for faster and more efficient results
  • Reduces dependency on backend infrastructure by shifting computation to the client side, improving performance and responsiveness
  • Eliminates the need for maintaining heavy ML servers, helping reduce operational costs while ensuring scalable performance

Cost-Effective ML Validation
icon Incrementally Improving Performance

  • Built using the YOLO v2 real-time object detection model to support accurate and efficient license plate detection in dynamic environments
  • Detects small objects within large scenes effectively while maintaining accuracy in complex and high-density visual conditions
  • Uses a high-resolution classifier to improve localization accuracy and ensure precise detection of license plates across frames
  • Processes the entire image in a single neural network pass, reducing computation time and improving overall system performance

Incrementally Improving Performance

The Impact

With Daffodil’s AI expertise, the client enabled law enforcement agencies to use their existing infrastructure for more proactive policing. The ML-based ALPR system allowed officers to act on historical leads from the National Crime Information Center (NCIC) and streamline investigations. The solution has scanned over 80,000 vehicles with an accuracy rate of 98.5%, improving reliability in real-world conditions.

The system also enabled faster follow-ups on long-pending cases by improving access to relevant data and insights. With 10x faster scanning and reporting, law enforcement teams could respond more efficiently in the field. The client acknowledged Daffodil’s ability to deliver results that aligned well with their operational expectations.

icon

80,000+

Vehicles Scanned

icon

98.5%

Accuracy

icon

10x

Faster Scanning & Reporting

Read Related Case Studies

Enabling India’s leading multi-brand online beauty retailer to leverage AI and achieve a 40% add-to-bag conversion rate

Daffodil helps an urban planning firm to leverage AI and detect green space from geo-spatial data.

Reserve Bank of India (RBI) helps the visually impaired to identify Indian banknotes denomination through an AI-enabled app.