JIVISIONJISHIKEJI
Get a Quote
中 EN
Home / Case Studies / Multi-camera Personnel Positioning
Project Case / Multi-camera Vision

Multi-Camera Personnel Positioning PlatformRTSP ingest, site-map calibration
and trajectory visualization

This case uses eight IP cameras as the reference configuration and brings RTSP ingest, installation registration, ground calibration, person detection, coordinate mapping, trajectories and alarms into one web console, with configuration-driven channels and split compute tasks for planned expansion.

Multi-stream RTSP ingestYOLOX + motionHomographyTracks and alarmsEngineering solution
View Solution DetailsAssess a Similar Project →
Multi-streamconfiguration-driven ingest
Unified mapsite-coordinate mapping
Track managementreplay and zone events
REST / WSsystem interfaces
SITE MAP CAM-01CAM-02CAM-03CAM-04CAM-05CAM-06CAM-07CAM-08 MULTI-CAMERA POSITIONING PLATFORM RTSP · CALIBRATION · FUSION · TRACK
Multi-camera personnel-positioning platform topology

Project overview

This case addresses personnel positioning and trajectory management across multiple fixed IP cameras, using eight cameras as the reference configuration. The platform connects RTSP ingest, person detection, ground-coordinate mapping, site-map display, trajectory replay and event integration in one workflow.

Solution objective: unify person observations from separate camera views in one site coordinate system while retaining an expansion path through configuration-driven camera registration, independent ingest tasks and consistent data interfaces.

Core modules

  • Configuration-driven multi-stream RTSP ingest and independent frame caches
  • Camera online state, frame-age and reconnect management
  • Site map, installation points and coverage configuration
  • Per-camera ground calibration and homography mapping
  • Person detection, foot-point extraction and local tracking
  • Cross-camera spatial association and track management
  • Live map, stream switching, replay and event views
  • REST and WebSocket business interfaces

Site map and camera topology

The reference site drawing defines an approximately 6.65 m × 6.20 m coordinate range. Eight physical camera positions and mounting heights are registered on the map. Every camera maintains its own mapping into the shared site coordinates, and an added viewpoint follows the same configuration and calibration flow.

System Flow

From eight video streams to a site-map position

Video ingest, person detection, local tracking, ground mapping, cross-camera association and business output form one processing chain.

RTSP IngestEight sub-stream workers
DetectionMotion candidates + YOLOX
Local TrackBox smoothing and foot point
Ground MapFour-point homography
FusionClusters, priors and outliers
OutputPositions, tracks and alarms
Definition and Scope

What is a multi-camera personnel-positioning platform?

Direct answer

A multi-camera personnel-positioning platform collects video from fixed cameras, detects and tracks people in each view, and uses ground calibration to convert image coordinates into one site-map coordinate system. It provides positions, trajectories, zone events and camera-status data. Compared with ordinary CCTV, it adds unified coordinates, cross-camera association, track management and business-system interfaces.

Capability layerMain outputProject configuration
Video monitoringMulti-view video, recording and playbackCamera connectivity, codec and storage policy
Person detection and single-view trackingPerson targets, local tracks and detection statesView, image quality, occlusion and model configuration
Multi-camera site-map positioningUnified map coordinates, cross-area tracks and eventsGround calibration, overlap, time alignment and business rules

Application scenarios and design considerations

Typical applications

Fixed-camera factory aisles, warehouse work zones, campus areas, laboratories and robot-collaboration zones can use the platform for occupancy, zone entry, trajectory replay and event integration.

Complex-scene adaptation

Dense crowds, occlusion, backlight, night scenes, reflective floors and camera-overlap areas are addressed through site samples, camera planning, image settings and track-association strategies.

Multi-level spaces

Stairs, ramps, multi-level platforms and areas with significant height changes use zone-specific calibration, multiple ground planes or a 3D positioning method so the coordinate model matches the site.

Identity and privacy

The primary output is person position and trajectory state. Named identity can be integrated through badges, authorized identity systems or another compliant method, together with access control, retention policies and audit logs.

Core Capabilities

Platform capability coverage

The platform connects multi-stream video, unified coordinates, trajectory management, scalable scheduling and system integration in one data-processing chain.

Multi-streamConfiguration-driven ingest and independent status management
Unified mapCalibrated mapping from image coordinates to site coordinates
Track managementCross-camera association, replay and zone events
REST / WSBusiness-system data interfaces and live status updates
CapabilitySolution contentApplication output
Video ingestMultiple RTSP sub-streams, camera registry, independent workers, frame caches and connection statesLive video, camera status and maintenance data
Positioning and fusionPerson detection, foot-point extraction, ground mapping, spatial clustering and cross-camera associationSite-map positions, source cameras and association states
Tracks and eventsTrack smoothing, zone entry/exit, dwell, boundary crossing and replayTrajectory data, zone events and alarms
Platform and interfacesWeb console, REST APIs, WebSocket updates and access configurationVisual management and business-system integration
Key Technologies

Technical design for multi-stream video positioning

Multi-stream ingest and status management

Each RTSP stream uses an independent worker and frame cache. Online state, frame age, reconnect count and status codes are managed consistently for scheduling and maintenance.

Browser video switching

The active video connection is released before the target channel is opened. Status updates are separated from the video component to manage multi-stream previews and browser connections.

Fixed-period scheduling

Detection and status updates use a fixed total-period scheduler, while frontend synchronization prevents overlapping requests and keeps multi-stream tasks aligned.

Track stability and cross-camera association

Foot points, spatial distance, time continuity, camera topology and track states are combined for association, position smoothing and state management.

Scalable Architecture

How the eight-camera reference configuration can expand

Direct answer

This case uses eight fixed IP cameras as the reference configuration. The platform is organized around a camera registry, independent ingest tasks, per-camera calibration records and consistent service interfaces. Additional cameras can reuse the site-map, trajectory and event data models after channel configuration, resource allocation, per-camera calibration and overlap-zone association. Practical capacity depends on codec settings, network bandwidth, decode and model workload, CPU/GPU resources, storage policy and refresh targets, so each expansion requires capacity planning and acceptance at the intended scale.

Configuration-driven camera registry

Each camera_id stores its RTSP settings, codec parameters, mounting point, coverage, enable state and calibration version, allowing channels to be added through configuration.

Independent ingest and state isolation

Each stream has its own ingest task, frame cache and runtime state. Ingest, decoding and inference can be split by process or compute node as channel count grows.

Per-camera calibration and zone association

Every added camera keeps its own ground mapping and is reviewed in overlap zones before joining the unified site coordinates and trajectory-association flow.

Consistent data and service interfaces

Positions, trajectories, events and camera health use stable data objects, exposed through the same REST or WebSocket interface pattern.

Capacity and acceptance checks before expansion

CheckPlanning basisExpansion action
Network and videoPer-stream bitrate, resolution, frame rate, keyframe interval and channel countReserve link headroom and configure sub-stream and reconnect policies
Decode and inferenceCodec, model workload, processing period and concurrent tasksAdd workers, processes or compute nodes and reconfirm scheduling targets
CPU, GPU and memoryPer-channel resource use, peak concurrency and cache policySize resources for the intended scale and run load and stability validation
Storage and dataRecording policy, track retention, event volume and logsAdjust capacity, retention rules and archive policy
Calibration and associationNew viewpoints, coverage boundaries, overlaps and site geometryComplete per-camera calibration, overlap review and expansion acceptance
Implementation Method

Six steps from site inputs to project delivery

1. Requirements and site modelConfirm drawing scale, coordinate origin, target areas, restricted zones, target types, trajectories and alarm rules.
2. Video-path configurationConfirm RTSP codec, resolution, frame rate, keyframe interval, clock alignment, reconnect policy and network resources.
3. Camera and coverage registryRecord mounting point, height, direction and overlap, then configure blind zones, occlusion areas and multi-plane regions.
4. Ground calibrationCollect distributed ground points for every camera, build a homography and review the mapping at independent and overlap points.
5. Detection, fusion and integrationConfigure person detection, foot points, candidate clustering, outlier rejection, track smoothing and cross-camera states using site samples.
6. Delivery and acceptanceUse project targets to define routes, samples, lighting, person counts and metric rules, then accept positions, tracks, events and interfaces.
Interfaces and Data

What the platform can provide to a business system

The platform uses a FastAPI backend and web console for status queries, live data updates and system integration. Fields, authentication, timestamps, coordinate units, state enums and retention policies can be defined for each project.

Data objectTypical fieldsBusiness use
Camera healthOnline state, frame age, reconnect count and status codeVideo-path management and maintenance alerts
Person observationSource camera, box, foot point, timestamp and association stateSingle-view review and traceable positioning data
Site-map positionMap X/Y, unit, calibration version and source-camera setArea logic, map display and upper-layer rules
Tracks and eventsTrack points, entry/exit, dwell, boundary crossing and alarm stateReplay, statistics and business-system integration
Project input checklist: scaled site drawing, camera models and sub-stream settings, mounting points and heights, sample video, target and occlusion conditions, expected error and refresh period, alarm rules, interface protocol, server environment, and data/privacy requirements.
Calibration and Delivery

Site-map calibration and project acceptance

Unified coordinates and acceptance

Each camera uses distributed ground points to build a homography, followed by mapping checks at independent points and overlap areas. Accuracy, refresh period, trajectory continuity and event response are defined for the target application.

  • Ground points cover the main positioning area, with coordinate units and calibration versions recorded.
  • Single-camera and overlap areas use separate measurement points and statistical rules.
  • Position deviation, track continuity, event response and interface periods follow business requirements.
  • Calibration is updated after changes to camera angle, focal length, resolution or mounting position.
  • Stairs, ramps and multi-level platforms use zone-specific, multi-plane or 3D coordinate models.
Project Delivery

Platform software and project deliverables

Backend servicesCamera, calibration, detection, association, trajectory, site-map and WebSocket services.
Web consoleSystem overview, live positioning, camera management, tracks, events, replay and settings.
Calibration and configuration toolsSite map, installation points, ground coordinates, zone rules and camera settings.
Interfaces and project documentsInterface fields, deployment configuration, operating instructions, calibration records and acceptance criteria.
FAQ

Multi-camera personnel-positioning questions

How does a multi-camera personnel-positioning platform work?

The platform reads multiple RTSP streams, detects people and extracts ground-reference points in each view, maps those observations into one site coordinate system, and uses spatial consistency and track state to report either candidate or fusion-confirmed positions.

How is it different from CCTV or person detection?

CCTV primarily provides video and recording, while person detection reports boxes in each view. Site-map positioning additionally converts pixels into unified coordinates and manages calibration states, cross-camera observations, tracks and events.

Why is ground calibration still required after cameras are registered on the drawing?

An installation point shows where a camera is mounted but does not convert image pixels into real ground coordinates. Each camera still needs a homography built from distributed known ground points, followed by overlap-residual checks.

How is personnel-positioning accuracy defined?

Accuracy is defined against the target site's map, calibration points and business zones. Acceptance metrics can be specified separately for single-camera areas, overlap areas and trajectory continuity, together with measurement points, statistical rules and scene conditions.

Does multi-camera positioning always require face recognition or ReID?

Not necessarily. Fixed sites can use topology, time, spatial distance and track continuity for cross-camera association. ReID, badges or an authorized identity system can be added when named identity or long-term identity continuity is required.

How is the platform refresh period determined?

The refresh period depends on camera encoding, network transport, video decoding, model workload, stream concurrency and business interfaces. A project can set a target period for track display, zone events or real-time integration and include it in acceptance criteria.

Can an existing IP-camera system be reused?

Cameras that provide stable RTSP sub-streams can be considered for reuse. Codec, resolution, frame rate, keyframe interval, mounting view, clock alignment and network resources are configured for the project.

Can the platform expand beyond eight cameras?

Yes, for a defined target scale. The platform uses a camera registry, independent ingest tasks, calibration records and consistent interfaces. Added channels require capacity planning, per-camera calibration, overlap-zone association, and performance and stability acceptance at the intended scale.

Can position data connect to MES, WMS or a campus platform?

REST or WebSocket interfaces can provide camera status, person observations, site-map coordinates, tracks and alarm events. Integration defines fields, authentication, timestamps, units, state enums and retention policies.

Project Inquiry

Assess a multi-camera positioning and trajectory project

Please provide the site drawing, camera count and parameters, mounting locations, target type, sample video, coverage overlaps, expected accuracy, refresh latency and system interfaces.