get in touch

Computer Vision Solution for Effective Advertising Placement

computer vision device

Computer Vision Solution for Effective Advertising Placement

Our team developed an on-premises device based on AI technologies for detecting individuals and showcasing advertisements on DOOH displays in transportation or outdoor locations.

#AI

#WebDevelopment

#Ecommerce

Client*

The client is a world-leading provider of comprehensive visual technology solutions.

*We cannot provide any information about the client or specifics of the case study due to non-disclosure agreement (NDA) restrictions.

Project in numbers

duration

11 months

team

5 specialists

Team involved in the project

industry

E-commerce

solution

Computer vision and artificial intelligence device for target audience analysis

technologies

Python, Pytorch, Cnvrg.io, AWS SageMaker, GCP Vertex AI, Fastdup, Pandas, Numpy, Scipy.

1 x Lead Data Scientist

2 x Data Scientists

2 x Data Engineers

HIRE THE TEAM

Challenge

The fundamental idea was to create a device harnessing AI technologies for analyzing captured images, discerning the audience’s average attributes, and enabling the presentation of targeted advertising.

Related objectives

Deploy the computer vision model on edge devices with limited GPU and RAM

Train the AI computer vision model with labeled data

Solution & functionality

In collaboration with the client’s team, Timspark created a compact on-premises computer vision device that can capture and analyze visual data.

Machine learning model training for successful object recognition

The team developed and fine-tuned the classifier model to meet the project’s unique requirements. The model analyzes received images, determines the average class among all attributes of captured objects, and subsequently identifies the target audience for the advertisement.

Deployment on Jetson Nano and Jetson Xavier

To address challenges linked to the limited memory and slow data processing, the model was transformed into ONNX and TensorRT formats, ensuring its seamless deployment on edge devices such as Jetson Nano and Jetson Xavier.

Results and business value

Our specialists successfully developed a classifying model that detects and tracks individuals within the camera’s field of view. The model analyzes the captured visual data, accurately identifying selected class among all attributes, and displays targeted advertisements on nearby screens in accordance with the specified class request.

Benefits for client

The client successfully applies the computer vision software to show the advertisement effectively to the relevant audience.

Based on the successful results, we are going to improve the model further in our ongoing collaboration.

Related cases

see all case studies

We appreciate your interest in Timspark

    ML in banking: 5 times less fraud risk by spotting weird transactions

    Machine learning in banking
    ML in banking: 5 times less fraud risk by spotting weird transactions
    Timspark leveraged the power of machine learning in banking to keep an eye on digital transactions and catch any abnormal behavior with a new extension for the existing client’s system.
    #Fintech #Banking
    #MachineLearning
    #DataAnalytics

    Client*

    We have partnered with a major bank that has branches all over the US, providing loans, deposits, and more banking products.
    *We cannot provide any information about the client or specifics of the case study due to non-disclosure agreement (NDA) restrictions.

    Project in numbers

    duration
    19 months
    team
    13 specialists

    Team involved in the project

    industry
    Banking, Fintech
    solution
    ML-based system for fraud risk analysis and detection
    technologies

    Python, Scala, DVC, MLFlow, Comet, Apache Spark MLLib, Scikit-learn, LightGBM, XGBoost, Hyperopt, PySpark, Numpy, Pandas, Scipy, Docker, Docker Compose, Kubernetes, Jenkins

    2 x Frontend Developers
    2 x Backend Developers
    1 x Project Manager
    1 x Business Analyst
    2 x Data Engineers
    3 x ML Engineers
    1 x QA Engineer
    1 x UX/UI Designer
    HIRE THE TEAM

    Challenge

    The key American bank faced rising financial fraud threats, and traditional systems proved ineffective. We were picking the best ways to use machine learning in banking and finance against increasing fraudulent activities that endangered customer safety and the bank’s reputation.

    Related objectives

    Implement ML for fraud detection
    Upgrade the anti-money laundering system
    Increase customer safety
    Improve the bank’s reputation

    Solution & functionality

    We suggested adding an ML-powered extension to the banking system to scrutinize large data volumes and protect funds from malicious activities. It analyzes account holders’ transactions and raises alerts for any unusual, suspicious, or fraudulent behavior. With deep learning fintech algorithms, our team processed extensive data to spot irregularities signaling potential fraud risk.

    Aggregating data

    To begin, our engineers collected and unified all banking data, encompassing user identities, transaction histories, locations, payment methods, and other pertinent factors.

    Detecting anomalies

    We identified distinctive patterns like high transaction amounts or segmented transactions to avoid automated tax reporting, enabling ML algorithms to distinguish fraud from regular banking. Transactions are tagged as “good” or “bad”.
    We also accessed a vast dataset, efficiently spotting patterns and anomalies, and selected crucial features through data comparison and elimination techniques, improving fraud risk analysis and detection.

    Training the ML model

    Our ML team created algorithms to catch odd situations that slip past regular rules. This extension can predict even with less data, using smart machine-learning methods. So, our solution uses embedded representations, not typical features, to handle transactions.

    Implementing the ML model

    Once a threat’s spotted, the system shoots real-time data to the admin, who can stop or nix operations for further digging. Depending on the fraud chance, there are three outcomes:
    • If fraud odds are below 5%, the transaction gets the green light.
    • If the odds range between 6% and 70%, an extra check like an SMS code, fingerprint, or secret question is needed.
    • If the fraud chance tops 80%, the transaction’s axed, needing hands-on analysis.
    Plus, we set up good ML tools to explain models, making predictions clear and keeping things smooth for users.

    Results and business value

    Timspark’s top-notch ML extension spots fraud and takes action. Security’s solid — no breaches or financial crimes.

    x2.4 speedier in processing

    Our ML algorithms swiftly handle heaps of data, keeping up with the rapid transactions.

    99.3% accuracy of fraud detection

    Using these algorithms, we find tricky patterns that humans might miss. That means fewer mistakes and less unseen fraud.

    Less mundane tasks

    Our solution checks hundreds of thousands of payments per second, making the transaction process as painless as possible.
    The algorithms catch tiny changes fast, checking tons of payments per second. The bank gets tighter security, faster transactions, and less chance of missed fraud. It means smoother banking and peace of mind for the end customers.

    Related cases

    see all case studies

    We appreciate your interest in Timspark

      Let’s build something great together

        Let’s build something great together

          Let’s build something great together