Computer Vision Solution for Secure Remote Work Environments

US-based technology-enabled global business services company specializing in customer engagement and business performance. 

The core challenges involved applying intensive, large-scale video processing to remote, dispersed environments securely and affordably:

• Massive Scalability: Processing hundreds of thousands of images for face recognition and object detection required massive computational power and efficient processing.
• Security & Compliance: The system needed to securely monitor unauthorized individuals or blacklisted objects (mobile devices, papers) to maintain expected data security and compliance standards.
• Cost Prohibitions: Initial image processing costs were prohibitively high, threatening the economic viability of continuous, large-scale monitoring.
  Verification Fraud: Required a robust anti-spoofing mechanism to prevent false facial verification using photos or videos.

We developed an AI/ML-powered computer vision solution that provides secure, real-time remote workspace monitoring. The solution detects, tracks, and verifies human faces in real-time with the help of deep learning neural network architecture. The anti-spoofing model prevents false facial verification by using a substitute (photo, video, mask) of an authorized person’s face. The solution is powered by TensorFlow for deep learning capabilities and deployed using Docker for massive-scale computational efficiency.

The solution ensures security and efficiency through:

• Real-Time Facial Recognition: Utilized deep learning neural networks to detect, track, and verify human faces in real-time. The anti-spoofing model prevents false facial verification.
• Object Detection Engine: The system detects objects or devices that are not permitted in the workspace, preventing unauthorized data capture and rule violations.
• Optimization for Scale: Implemented a comprehensive optimization strategy leveraging computationally efficient architecture, including batching, threading, and model runtimes (TensorRT, ONNX) to handle image processing on a massive scale.
• Scalable Deployment: Established a robust deployment environment using Docker containers and Kubernetes to collaborate and scale workloads on a cluster.
   

• Computational Efficiency: Achieved high performance using techniques like batching, threading, and scheduling operations between CPU and GPU, vital for large-scale face recognition.
• Optimized Deployment: The deployment environment using Docker containers and Kubernetes was crucial for cost-efficiency and scaling workloads.
• Core Technologies: Built on .NET and Angular for the application stack, leveraging AWS for cloud services and TensorFlow for all AI/ML models.
• Verification Logic: Verification checks are performed securely during login and at consistent, regular intervals throughout the employee's shift.

• Massive Cost Reduction: Advanced optimization using the Docker-container architecture resulted in a 98.75% reduction of GPU utilization cost (from $1,600 to $20 per frame) for image processing.
• Increased Productivity: Computer vision aided monitoring reduced workspace distractions, resulting in a 45% increase in measured productivity after implementation.
• Enhanced Risk Management: The high level of visibility and automated violation alerts enhanced compliance and reduced data security risks by an estimated 55%.
• Flexibility and Scalability: The containerized architecture provided the ability to operate 24x7 and scale up or down based on demand with high operational flexibility.