Skip to content
VIDEO.MA
FR

Solutions

The video analytics building blocks of the VIDEO.MA platform

Each capability is presented with how it works, the problem it solves, its technical approach, its metrics and its limitations. They can be combined to meet your needs.

Request a demo

Video data vs. business data: an essential distinction

Video counting measures passages, presence or durations. It does not replace your POS, inventory or sensor data. When that data is needed (actual revenue, stock levels, consumption), it comes from an integration with your systems. Every metric is labeled according to its source.

LegendPer projectCapability deployed project by project, following a feasibility study.Feasibility studyHighly dependent on the site and integrations: subject to a dedicated study.Coming soonPlanned for the future VIDEO.MA platform, not available today.

People counting

Automatic, anonymous counting of entries, exits and footfall using your cameras.

Per projectVideo data

What it does

People counting automatically measures visitor flows through an entrance, a corridor or a virtual line drawn on the image. Crossings are counted by direction (in / out) and aggregated by time slot, without identifying anyone.

Problem solved

Manual counts are occasional, costly and unreliable. Without continuous footfall data, it is hard to calculate conversion rates, size your teams or compare activity across sites and periods.

Technical approach

Video streams are pulled over RTSP / ONVIF from existing or new cameras and analysed locally on an Edge AI box. A person detection model, combined with multi-object tracking, gives each silhouette a temporary anonymous track; crossing a virtual line triggers a timestamped event. Only aggregated events are sent to the database and reports.

Measurable metrics

  • Entries and exits per hour, day and week
  • Peak hours and peak days
  • Footfall comparison between sites or periods
  • Conversion rate (combined with integrated POS transactions)

Limitations and prerequisites

Accuracy depends on camera height and angle (an overhead view works best), lighting, resolution and occlusion when people pass in dense groups. Virtual lines need calibration, and actual accuracy must be measured on site during a pilot against a reference count.

Vehicle counting

Counting vehicles entering, leaving or passing through a lane, bay or car park.

Per projectVideo data

What it does

Vehicle counting detects and counts cars, vans, trucks or two-wheelers crossing a virtual line or entering a zone. Where the camera view allows, data is broken down by vehicle type and aggregated by period.

Problem solved

Car park, fuel station, car wash and logistics operators often lack reliable data on real traffic: volumes per time slot, peaks and vehicle mix. Yet this information is essential for sizing staff and equipment.

Technical approach

Cameras are connected over RTSP / ONVIF to an Edge AI computer running a vehicle detection model and a multi-object tracking algorithm. Virtual lines or zones determine crossings by direction of travel; events are aggregated and exposed through APIs and reports. No licence plate reading is performed by default.

Measurable metrics

  • Vehicles per hour and per direction
  • Breakdown by vehicle category (depending on the camera view)
  • Traffic peaks and quiet periods
  • Volume per lane, bay or entrance

Limitations and prerequisites

Performance varies with viewing angle, distance, vehicle speed, headlights at night, rain or dust, and vehicles hiding one another. Fine-grained classification by type requires sufficient resolution and framing. Accuracy must be validated on site during a pilot.

Object detection

Automatically locating people, vehicles or business-specific objects in the image.

Per projectVideo data

What it does

Object detection locates items of interest in each frame — people, vehicles, trolleys, pallets, containers — and assigns each one a category and a confidence score. It is the building block of most Smart Vision analytics.

Problem solved

A conventional camera records images but does not “understand” what it sees. Object detection turns that passive stream into usable data: the presence, number and position of the items that matter to your business.

Technical approach

Convolutional neural network detection models (YOLO family or equivalent, with licensing reviewed for each project) run inference on the Edge AI computer from RTSP streams. For objects specific to your business, retraining on annotated images from your site can be assessed. Confidence thresholds are tuned during the pilot.

Measurable metrics

  • Number of detected objects by category and zone
  • Presence / absence of an object type over a period
  • Detection rate measured against a reference sample during the pilot

Limitations and prerequisites

Small, distant, partially hidden or poorly lit objects are harder to detect. Non-standard objects often require an annotated dataset and specific training, to be assessed during the feasibility study. Accuracy is never guaranteed upfront: it is measured on site.

Object tracking

Following objects and people from frame to frame, anonymously.

Per projectVideo data

What it does

Object tracking links successive detections of the same item to reconstruct its path within the camera’s field of view. It makes it possible to measure paths, dwell times and line crossings using temporary, anonymous identifiers.

Problem solved

Counting detections frame by frame is not enough: the same customer or vehicle would be counted many times. Tracking is what enables reliable counting, duration measurement and path analysis within a space.

Technical approach

A multi-object tracking algorithm (ByteTrack or equivalent) associates detections across frames by combining position, predicted motion and confidence score. Each object receives a temporary ID, valid only within a single camera’s view and unrelated to anyone’s real identity. Tracks are used to generate events (zone entry, line crossing, duration).

Measurable metrics

  • Aggregated paths and busiest areas
  • Average dwell time per zone
  • Line crossings per virtual line and direction

Limitations and prerequisites

Long occlusions, dense crowds or sudden lighting changes can cause lost tracks or ID switches. Tracking is limited to each camera’s field of view: following a person across several cameras is not offered by default, notably for data protection reasons (Moroccan Law 09-08).

Zone monitoring

Defining virtual zones and measuring the activity that takes place in them.

Per projectVideo data

What it does

Zone monitoring lets you draw areas of interest on the image — an aisle, a checkout, a loading dock, a wash bay, a restricted area — and automatically measure entries, exits, presence and dwell time in each.

Problem solved

Operations managers need to know which areas are actually used, saturated or neglected, and whether sensitive areas are occupied outside expected hours.

Technical approach

Virtual polygons are configured on each camera stream. Objects detected and tracked on the Edge AI box are matched against these polygons to produce events (entry, exit, prolonged presence, out-of-hours presence). Rules can be set per zone and per time window, then aggregated in the database.

Measurable metrics

  • Entries per zone and period
  • Average dwell time per zone
  • Zone utilisation rate throughout the day
  • Out-of-hours presences

Limitations and prerequisites

A zone is only reliable if it is clearly visible to the camera, without obstacles or significant overlap. Strong perspective makes precise ground-level boundaries harder to draw. Zones must be recalibrated whenever a camera is moved.

Occupancy analysis

Real-time occupancy of a space, room or car park.

Per projectVideo data

What it does

Occupancy analysis estimates how many people or vehicles are present in a space at any moment and compares it with capacity. Use it to monitor how full a dining room, terrace, office floor or car park is.

Problem solved

Without occupancy data, spaces are sized on intuition: underused floor space, unanticipated overcrowding and maximum capacities that are hard to enforce.

Technical approach

Two methods are combined depending on the site: directly counting the people or vehicles visible in a zone, or balancing entries and exits at access points. Measurements are computed on the Edge AI box, smoothed over time and compared with the configured capacity to produce rates and thresholds.

Measurable metrics

  • Live and average occupancy rate
  • Daily peak occupancy
  • Time spent above a capacity threshold
  • Free / occupied parking spaces per zone

Limitations and prerequisites

Entry / exit balancing can drift during the day if counting errors accumulate; a periodic reset is built in. Large spaces need several cameras and overlap handling. Measured occupancy is an estimate whose margin of error must be assessed during the pilot.

Queue detection

Measuring queue length and alerting when waiting becomes excessive.

Per projectVideo data

What it does

Queue detection measures how many people or vehicles are waiting at a checkout, counter, service desk or wash bay, and estimates waiting time.

Problem solved

Long queues damage the customer experience and cost sales, while opening an extra till often happens too late. Managers lack an objective measure to react and plan.

Technical approach

A queue zone is defined on the image. The Edge AI box continuously counts the people or vehicles in that zone and, using multi-object tracking, measures the time from joining the queue to reaching the service point. Configurable thresholds trigger notifications.

Measurable metrics

  • Average and maximum queue length
  • Estimated average waiting time
  • Threshold breaches per day
  • Queue length versus number of open service points

Limitations and prerequisites

Non-linear queues, customers waiting outside the zone or accompanying people can skew measurements. A dedicated camera mounted high and facing the queue usually performs better than an existing security camera. Thresholds must be tuned on site.

Service duration

Measuring time spent at a counter, in a bay or in a service area.

Per projectVideo data

What it does

Service duration measurement calculates how long a person or vehicle spends in a service area: checkout, order taking, washing, refuelling, dock loading.

Problem solved

Service times directly determine a site’s capacity. Without measurement, you cannot identify bottlenecks, compare service points or check the impact of a reorganisation.

Technical approach

Each object tracked in a service zone gets an entry and exit timestamp. The Edge AI box computes the duration, filters out very short passes (people simply walking through) and aggregates results by service point, zone and period. Durations can then be matched with other business data through integration.

Measurable metrics

  • Average and median service time per service point
  • Distribution of durations (short, long, very long)
  • Services completed per hour
  • Change in durations after an operational change

Limitations and prerequisites

The measured duration reflects presence in a zone, not necessarily actual service time: atypical behaviour (waiting on the spot, back-and-forth movement) can create gaps. Zone definitions and filtering rules are validated during the pilot.

Process productivity monitoring

Throughput and utilisation of workstations, docks and bays — not individuals.

Feasibility studyVideo data

What it does

Productivity monitoring measures process activity: cycles completed at a workstation, utilisation of a loading dock or wash bay, idle time in a production area. The analysis focuses on processes and equipment, never on monitoring individual employees.

Problem solved

Many businesses have no objective measure of the real throughput of their operational areas. They struggle to identify idle time, workload imbalances between stations and unused capacity.

Technical approach

Virtual zones are mapped to workstations, bays or docks. The Edge AI box detects relevant objects (vehicles, trolleys, pallets, products) and derives activity states (busy, free, in cycle) from configured rules. Indicators are aggregated by zone and period, without identifying operators.

Measurable metrics

  • Utilisation rate per workstation, dock or bay
  • Cycles or passes per hour
  • Idle time per zone
  • Throughput comparison between zones and shifts

Limitations and prerequisites

What counts as a “cycle” depends heavily on each process and requires a dedicated feasibility study. Any deployment must comply with labour law and Moroccan Law 09-08, with employees informed beforehand and, where applicable, a declaration to or authorisation from the CNDP. VIDEO.MA does not offer individual employee scoring.

Inventory movement monitoring

Detecting pallets, parcels or trolleys moving between storage areas.

Feasibility studyVideo + business data

What it does

Inventory movement monitoring detects and counts logistics items — pallets, parcels, trolleys, totes — passing through key points: receiving docks, warehouse doors, aisles and picking areas.

Problem solved

Discrepancies between physical flows and flows recorded in the information system are hard to spot. An independent visual count of movements can help flag inconsistencies and better understand activity in logistics areas.

Technical approach

Detection models, adapted to the site’s objects where needed, run at the edge on RTSP streams from cameras covering key passage points. Multi-object tracking and virtual lines generate timestamped movement events, which can be reconciled with movements recorded in the WMS or ERP via API or data export.

Measurable metrics

  • Pallets or parcels per checkpoint
  • Dock activity per hour
  • Gaps between detected and recorded movements (with integration)

Limitations and prerequisites

Video does not read product references or the quantities inside a parcel: it detects objects, not SKUs. Reliability depends heavily on object shape, stacking, speed and framing. Reconciling with stock requires API or export access to the WMS / ERP. This capability is subject to a case-by-case feasibility study.

Potential stockout detection

A visual signal of empty shelf areas, to be confirmed by stock data.

Feasibility studyVideo + business data

What it does

Potential stockout detection visually identifies shelf areas that appear empty or depleted and sends a signal to replenishment staff. It is a visual cue, not a stock measurement.

Problem solved

An empty shelf loses sales, often without the store noticing quickly. Manual visual checks are irregular, especially at busy times.

Technical approach

A fixed camera frames a section of shelving. Shelf zones are defined and compared with a “fully stocked” reference state using detection or segmentation models run at regular intervals on the Edge AI box. A signal is raised when the empty ratio exceeds a threshold; it can be cross-checked with stock data through integration.

Measurable metrics

  • Empty-zone signals per shelf and per day
  • How long a zone remains flagged as empty
  • Time from signal to replenishment (if the action is logged)

Limitations and prerequisites

A visually empty shelf does not necessarily mean a stockout (stock in the back room, deliberate facing); the signal must be confirmed by stock data. Shoppers blocking the view, reflections, planogram changes and oblique angles reduce reliability. It usually requires dedicated cameras and a specific feasibility study.

Anomaly detection

Flagging unusual activity against configured rules or a historical baseline.

Feasibility studyVideo data

What it does

Anomaly detection flags situations that deviate from configured rules or observed patterns: presence in a zone out of hours, an object left stationary for an unusually long time, unexpected crowding, an area idle in the middle of the day. It draws staff attention; it does not make automated decisions.

Problem solved

Operators cannot watch dozens of screens continuously. Highlighting the situations that deserve a check focuses human attention where it is actually needed.

Technical approach

Two complementary approaches are assessed: explicit rules (zones, schedules, durations, count thresholds) evaluated on the Edge AI box, and comparison of aggregated indicators with a historical baseline to spot statistical deviations. Each signal carries a timestamp and, where the project allows, a snapshot for human review.

Measurable metrics

  • Anomalies flagged by type and zone
  • Share of signals confirmed after human review
  • Average time to handle a signal

Limitations and prerequisites

An “anomaly” must be precisely defined for each site; unforeseen or subtle situations may go undetected, and false alerts are inevitable, especially at the start. The system supports staff but never replaces certified security, alarm or detection systems.

Real-time alerts

Automatic notifications when a threshold or business rule is crossed.

Per projectVideo data

What it does

Real-time alerts notify the right people when a configured event occurs: a queue that is too long, capacity exceeded, out-of-hours presence, an idle area, a vehicle parked for too long.

Problem solved

Indicators reviewed at the end of the day arrive too late to act on. An alert at the right moment lets you open another till, reassign a team member or check an area while the situation is still unfolding.

Technical approach

Rules are evaluated continuously on the Edge AI box from detection, tracking and zone events. When a condition is met (threshold, duration, time window), an event is sent to the server, which handles de-duplication, logging and delivery through the channels chosen for the project (email, webhook, API to an existing tool).

Measurable metrics

  • Alerts by type, zone and period
  • Delay between event and notification
  • Time for teams to respond

Limitations and prerequisites

Notification delay depends on network quality and the channels used. Poorly tuned thresholds produce either too many alerts or too few: adjustment during the pilot is essential. VIDEO.MA alerts are an operational aid and never replace certified fire, intrusion or security systems.

Reporting

Periodic reports consolidating your video indicators by site and period.

Per projectVideo data

What it does

Reporting consolidates the indicators produced by video analytics — footfall, occupancy, queues, durations — into periodic reports (daily, weekly, monthly) that can be compared across sites and periods.

Problem solved

Scattered, unconsolidated data cannot support decisions. Regular, consistent reporting gives management a clear view of activity and how it evolves.

Technical approach

Aggregated events are stored in a PostgreSQL database. Scheduled jobs compute indicators by site, zone and period and make them available as reports and exports (CSV, Excel) or via API, in the format agreed with the client during the project.

Measurable metrics

  • Consolidated indicators by site, zone and period
  • Period-on-period changes
  • Site rankings and comparisons
  • CSV / Excel exports of aggregated data

Limitations and prerequisites

A report is only as reliable as the indicators feeding it, which must be validated on site. Camera stream or network outages create gaps in the data, which are flagged as such. Sales or stock indicators only appear when integrated with your systems.

POS & ERP integration

Combining video indicators with your sales, transactions and business data.

Feasibility studyBusiness integration

What it does

POS and ERP integration matches video-derived indicators with actual data from your information system: receipts, revenue, stock movements, staff schedules. It enables combined indicators such as conversion rate.

Problem solved

Video measures footfall and activity, not sales or stock. Without combining it with business data, you cannot tell whether higher traffic turns into revenue, or whether an empty shelf is a genuine stockout.

Technical approach

Business data is retrieved through the APIs, scheduled exports or database connectors provided by the POS or ERP vendor. It is normalised, aligned to the same time slots and sites as the video indicators, and stored server-side to compute combined indicators.

Measurable metrics

  • Conversion rate (transactions / entries)
  • Average basket relative to footfall
  • Revenue per visitor and per hour
  • Gaps between video signals and stock data

Limitations and prerequisites

Integration depends entirely on data access: a documented API, automated export or vendor agreement. Some POS systems offer no usable access at all. The quality of combined indicators also depends on time synchronisation and source data quality. Each integration requires a specific study.

Business dashboards

Dashboards delivered per project today; a self-service SaaS platform is on the way.

Coming soonVideo + business data

What it does

Business dashboards present your Smart Vision indicators visually and at a glance: footfall, occupancy, queues, durations, alerts. Today they are designed and delivered as part of each project; a self-service SaaS platform, letting each client configure their own views, is in preparation (Phase 2).

Problem solved

Executives and site managers need a quick read on activity, without handling raw files or watching video feeds.

Technical approach

In current projects, dashboards are built from aggregated data stored in the database and tailored to the selected indicators. The upcoming VIDEO.MA SaaS platform (React, Spring Boot, PostgreSQL) is planned to offer configurable multi-site dashboards, user and permission management, and export APIs.

Measurable metrics

  • Key indicators per site and multi-site view
  • Hourly, daily and weekly trends
  • Alert and event history
  • Combined indicators (with POS / ERP integration)

Limitations and prerequisites

The self-service SaaS platform is not available yet: it is on the roadmap. Dashboards delivered today are project-specific. The indicators shown depend on the capabilities actually deployed and validated on site.

Custom

Need a combination of building blocks?

Tell us about your challenge: we assemble the right capabilities and verify their feasibility at your sites.

Request a custom solution

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.

Request a demo Request a tailored quote