Computer vision
Algorithms that interpret image content to extract structured information: presence, position and movement.
Technology
Smart Vision technology built around your business indicators
VIDEO.MA combines computer vision, Edge AI processing and data integration to turn video streams into anonymous business indicators. Here we present our technology building blocks transparently, including where each one actually stands today.
Technology areas
Object detection
Neural networks that locate people, vehicles and business-specific objects in each frame, with a confidence score.
Object tracking
Multi-object tracking that links detections across frames using temporary, anonymous identifiers.
Edge AI
Analysis runs on site, close to the cameras: less bandwidth, faster response, and no images transmitted by default.
Video analytics
Virtual lines and zones, rules and durations that turn detections into business events: counts, occupancy, queues.
IoT & sensors
Video can be complemented by sensors (counting, environment, equipment) when the use case calls for it, subject to a specific study.
Business intelligence
Aggregated indicators, trends and cross-site comparisons, presented in actionable reports and dashboards.
Data integration
Matching with your POS, ERP or WMS via API or export, to separate observed activity from actual sales and stock.
Real-time monitoring
Rules evaluated continuously and alerts sent to the relevant teams, supporting their operational decisions.
Our technology stack
The statuses below reflect our current implementation status: “in use” for the technologies running our current web platform, “evaluated” for components tested and selected for video analytics projects and the platform in preparation, and “planned” for those on our roadmap.
| Technology | Role | Status |
|---|---|---|
Spring Boot Back end Java framework powering the API and server-side services of the VIDEO.MA web platform. | Java framework powering the API and server-side services of the VIDEO.MA web platform. | In use |
PostgreSQL Database Relational database for the web platform, also selected for storing aggregated indicators. | Relational database for the web platform, also selected for storing aggregated indicators. | In use |
React / TypeScript Front end Interface of the VIDEO.MA website and foundation for the future client interface. | Interface of the VIDEO.MA website and foundation for the future client interface. | In use |
Python AI & video processing The reference language for computer vision pipelines, model training and evaluation. | The reference language for computer vision pipelines, model training and evaluation. | Under evaluation |
OpenCV Computer vision Image processing library: stream decoding, pre-processing, and geometry for virtual zones and lines. | Image processing library: stream decoding, pre-processing, and geometry for virtual zones and lines. | Under evaluation |
YOLO Object detection A family of real-time detection models. Model licensing terms are reviewed for each project. | A family of real-time detection models. Model licensing terms are reviewed for each project. | Under evaluation |
ByteTrack Multi-object tracking Multi-object tracking algorithm that links detections across frames anonymously. | Multi-object tracking algorithm that links detections across frames anonymously. | Under evaluation |
NVIDIA DeepStream Edge AI Hardware-accelerated multi-stream video processing framework for NVIDIA platforms, used for edge analytics. | Hardware-accelerated multi-stream video processing framework for NVIDIA platforms, used for edge analytics. | Under evaluation |
TensorRT Inference optimisation Model optimisation for fast inference on Edge AI computers equipped with NVIDIA GPUs. | Model optimisation for fast inference on Edge AI computers equipped with NVIDIA GPUs. | Under evaluation |
ONVIF Video standard IP camera interoperability standard, used to discover and configure existing equipment. | IP camera interoperability standard, used to discover and configure existing equipment. | Under evaluation |
RTSP Streaming protocol Protocol for accessing camera and recorder video streams, independent of any particular brand. | Protocol for accessing camera and recorder video streams, independent of any particular brand. | Under evaluation |
MQTT IoT & messaging Lightweight messaging protocol planned for collecting sensor data and IoT events. | Lightweight messaging protocol planned for collecting sensor data and IoT events. | Planned |
Our technical principles
Edge-first processing
Video streams are analysed on site by Edge AI computers. Only aggregated indicators are transmitted, reducing bandwidth and data exposure.
Privacy by design
Aggregated, anonymous analysis with no facial recognition or biometric identification. Projects are designed to comply with Moroccan Law 09-08 and CNDP requirements.
Accuracy measured, not promised
We never announce accuracy figures upfront. The reliability of each indicator is measured on site during a pilot, against a reference.
Open standards and existing infrastructure
RTSP and ONVIF make it possible to reuse compatible cameras and avoid lock-in to a single manufacturer.
System security
Video network segmentation, encrypted access, permission management and controlled updates of deployed equipment and software.
Frequently asked questions
Do you use facial recognition?
No. VIDEO.MA offers neither facial recognition nor biometric identification. Our analytics cover aggregated, anonymous counts, durations and occupancy.
Do images from my cameras leave my site?
Not by default. Analysis runs on site on an Edge AI computer and only aggregated indicators are transmitted. Any exception, such as sending a snapshot for verification, is agreed with you as part of the project.
How accurate is your analysis?
It depends on each site: camera positioning, lighting, traffic density. That is why accuracy is measured during a pilot, on your own data, before any large-scale rollout.
Is the SaaS platform available?
Not yet. The public website is live; the analytics engine and the SaaS client platform are in preparation (Phase 2). Projects are currently delivered case by case, following a feasibility study.
Can you integrate our sales or stock data?
Yes, provided your data can be accessed via API or export. Each integration is subject to a specific study, and video-derived indicators are always kept clearly separate from your sales and stock data.
Ready to turn your cameras into business intelligence?
Tell us about your business: we’ll show you what your cameras can measure, backed by an honest feasibility study.