Multi-Camera Human Safety Monitoring for Robotic Work Areas
Real-time human detection for high-risk industrial zones with per-camera ROI, event recording, alarm logic, health monitoring, watchdog behavior and PLC output integration.
AI Engineer • Industrial Computer Vision
Ashok Kumar Yadav
builds AI that works in the real world.
AI Engineer at Hesham Industrial Solutions, focused on production-grade computer vision, real-time video analytics, industrial safety systems, multi-camera AI, PLC integration, backend engineering and deployment-ready Windows applications.
Currently
Building production AI
AI Engineer • Hesham Industrial Solutions
7+Camera real-time AI systems designed for multi-stream operation
24×7Industrial monitoring and watchdog-oriented deployment mindset
AI + PLCComputer vision connected to industrial control workflows
Edge-firstGPU/CPU optimized Windows and local-network deployments
Selected work
Industrial AI systems built for actual operations.
My work sits at the intersection of computer vision, industrial safety, edge computing, backend systems and automation. The objective is not just model accuracy—it is reliable end-to-end operation.
DMS + PPE Multi-Camera Detection Platform
Driver monitoring and PPE compliance system supporting multiple cameras, configurable detection timers, per-camera settings, ROI, voice alerts, snapshots, statistics and hardware-aware GPU/CPU operation.
Real-Time Employee Recognition & Assembly Counting
Multi-camera face recognition workflow designed for gate and assembly scenarios with line-crossing events, employee enrollment, presence tracking, missing-person lists and large-population attendance logic.
Temporal Video Intelligence for Smoke & Fire Detection
Explored motion-aware video models and real-time inference pipelines to improve recognition of smoke-like temporal patterns while reducing false positives from static backgrounds.
PLC-Integrated Vision & Safety Control Workflows
Connected AI decisions to Siemens and Mitsubishi PLC logic using digital I/O, SSR interfaces, PROFINET-oriented workflows and safety-state logic for machine and operator interaction.
Production Backends, ERP & Deployment Systems
Built and refined Django/FastAPI applications, role-based workflows, PostgreSQL deployments, media storage, reporting, installers and Windows deployment pipelines for operational software.
Engineering approach
From model demo to deployable system.
The hardest part of industrial AI is rarely the model alone. Reliability, reconnect behavior, latency, storage, alarms, camera health and operator usability decide whether a system survives production.
Case study 01
Real-time safety monitoring across multiple RTSP cameras
Designed for continuous industrial operation where camera disconnects, frame lag, event logging and PLC signaling all have to be handled safely.
- Per-camera ROI and confidence configuration
- Local snapshots and short event video storage
- Camera reconnect, watchdog and system-health monitoring
- GPU-preferred inference with CPU fallback strategy
- PLC alarm integration and manual output testing
- Windows auto-start and service-oriented deployment
Case study 02
Reliable multi-camera DMS and PPE monitoring
Focused on reducing false alarms while keeping UI and detection behavior understandable for operators.
- Independent confirmation and cooldown timers by class
- Strict PPE violation handling and snapshot evidence
- Camera assignment, ROI and reconnect logic
- Voice alerts and event history
- Statistics and operational health views
- Installer-ready packaging and deployment fixes
Case study 03
Face recognition designed for moving groups
Worked on recognition logic for environments where several people can enter simultaneously and identity duplication must be controlled.
- Multi-angle enrollment workflow
- Line-crossing based arrival logic
- Recognition plus counting and present/missing status
- Multi-camera independence and reconnect handling
- GPU-oriented real-time performance
Technical stack
Tools are secondary. Systems thinking comes first.
I choose technologies around latency, reliability, maintainability and deployment constraints rather than building around a single framework.
Computer Vision
YOLO, real-time detection, tracking, face recognition, PPE, DMS, ROI logic, RTSP pipelines, temporal video models.
AI Deployment
NVIDIA GPU inference, CPU fallback, model optimization, multi-camera orchestration, PyInstaller, Windows services.
Backend
Python, FastAPI, Django, REST APIs, SQLite, PostgreSQL, authentication, role-based systems.
Industrial Integration
Siemens S7-1200, Mitsubishi PLC, PROFINET concepts, Modbus RTU, ESP32, SSR and digital I/O integration.
Reliability
Watchdogs, auto-reconnect, process recovery, storage management, health telemetry, logging and fault handling.
Frontend / UX
Responsive HTML/CSS/JS, dashboard UX, kiosk interfaces, camera grids, settings and operational workflows.
Deployment
Windows 11, installers, EXE packaging, local networks, Render, PostgreSQL and production configuration.
Technical Content
Founder of Xenors, writing practical content around AI, software, automation and technology.
Experience
Engineering through iteration, testing and deployment.
My focus has increasingly moved from software prototypes toward AI systems that must function continuously in industrial environments.
Current
AI Engineer — Hesham Industrial Solutions
Developing industrial computer-vision and automation systems including human safety monitoring, PPE/DMS detection, multi-camera AI, face recognition and PLC-connected applications.
Independent work
Founder & Technical Creator — Xenors
Building a technology knowledge platform focused on AI, software development, industrial automation, finance technology and practical technical education.
Engineering foundation
Computer Science & Backend Development
Developed practical experience with Python, Flask/Django, databases, authentication systems, backend architecture and full-stack product development while pursuing Computer Science Engineering.
About
I prefer systems that survive real-world conditions.
How I work
I start with the operational problem: what can fail, what must recover automatically, what the user needs to understand instantly, and what the hardware can realistically support. Then I design the AI, backend, UI and control logic around those constraints.
What I care about
Low latency, stable camera pipelines, accurate event logic, clean deployment, useful logging, predictable recovery behavior and interfaces that an operator can use without needing to understand the AI underneath.
Connect
Building something that needs AI to work outside the lab?
My strongest work is in computer vision, industrial safety, real-time video systems, AI-assisted automation and deployment-focused engineering.
