About Client
The client is a large-scale manufacturing organization focused on driving efficiency and productivity across its shop-floor environments. The company leverages technology driven analytics solutions to enhance operational visibility and enable real-time decision-making for machine operators and plant managers. With extensive manufacturing operations, the organization continuously processes machine and production data to reduce downtime, improve asset utilization, and optimize overall production performance.
The manufacturing organization has hundreds of cameras across its shopfloor that continuously creates a large volume of video data that requires significant compute, storage, and network resources to manage. While control room teams monitor these feeds, it becomes difficult to track everything manually due to the scale, leading to missed safety and operational incidents, especially when quick action is required. The organization needed a solution that could automatically analyze both real-time and stored video to detect key activities such as worker safety compliance, machine utilization, material movement, and unusual events.
However, implementing this comes with challenges. The system must process data quickly and generate alerts within minutes. Ensuring consistent results between live and historical video analysis adds further complexity. Additionally, too many alerts can overwhelm operators. Additionally, AI-generated summaries need to be carefully controlled to avoid incorrect insights.
To address these challenges, Comprinno implemented a scalable, production grade video intelligence platform built on a serverless, event-driven AWS architecture. The solution continuously processes feeds from cameras across the shopfloor, enabling real-time and retrospective video analysis to automatically detect and track critical operational parameters such as worker safety compliance, machine utilization, and zone intrusion.
The platform leverages ffmpeg, OpenCV, and purpose-built computer vision models (such as YOLO and DeepSORT) for deterministic, high-accuracy detection. Data summarization and insight generation are handled natively by Amazon QuickSight's built-in AI features (Amazon Q in QuickSight), eliminating the need for a separate LLM service. This architectural approach ensures reliable, explainable outputs for safety-critical use cases, with consistent performance across live and historical analysis.
Key Actions:
Foundation & Environment Setup:
Video Ingestion & Streaming Pipeline:
Computer Vision Detection Engine:
Data Storage & Management:
Data Extraction & Cataloging:
Visualization, Insights & Monitoring:
Comprinno designed a scalable and resilient architecture to enable real-time video analytics for manufacturing shop-floor operations. Video streams from IoT cameras (edge devices) are securely ingested into Amazon Kinesis Video Streams (KVS), which automatically provisions and scales infrastructure to handle high volumes of concurrent camera feeds. Within a secure Amazon VPC spanning three Availability Zones, containerized backend services running on Amazon ECS (Fargate) subscribe to KVS and execute a two-stage pipeline: frame extraction using ffmpeg and OpenCV, followed by YOLO inference for object detection and DeepSORT for object tracking across frames. Processed frames are stored in Amazon S3 as image files, while structured JSON detection data with S3 object keys is written to Amazon DocumentDB deployed across multiple AZs in dedicated DB subnets. AWS Glue Crawlers connect directly to DocumentDB within the VPC to discover schemas and populate the AWS Glue Data Catalog, enabling data discovery and ad-hoc querying via Amazon Athena.
For visualization and business intelligence, Amazon QuickSight provides interactive dashboards, charts, and drill-downs, with Amazon Q (built into QuickSight) enabling AI-powered summaries and natural language data exploration for plant managers and stakeholders. Monitoring and observability are handled using Amazon CloudWatch and AWS Systems Manager, while AWS CloudTrail ensures auditability. Security is enforced through AWS IAM, AWS KMS for encryption, AWS Certificate Manager for TLS/SSL management, and AWS Security Hub for continuous compliance and posture management.
By implementing this solution, Comprinno enabled the manufacturing operations organization to transform shopfloor monitoring from a manual and reactive process into an automated, insight-driven system. Continuous analysis of camera feeds improved visibility across safety, operations, and process efficiency, allowing teams to detect and respond to incidents in near real time while reducing dependency on manual supervision. The combination of deterministic computer vision models and a scalable AWS-based architecture ensured high detection accuracy, low-latency performance, and consistent outcomes across both real-time and historical analysis. By restricting the use of Amazon Bedrock to structured insight generation, the solution maintained reliability and eliminated risks associated with AI-driven misinterpretation in critical detection workflows.
Overall, the implementation delivered a robust and scalable foundation for intelligent shopfloor monitoring, enabling the organization to enhance safety compliance, optimize operational performance, and make faster, data-driven decisions. It also positions the customer to extend similar capabilities across additional manufacturing use cases, driving long-term efficiency and innovation at scale.
Comprinno Achieves AWS Data and Analytics Competency | Read More →
