English | 中文

Vision Hub Platform is an open-source platform for visual algorithm management, video stream access, task orchestration, frame capture, AI-powered image analysis, and event records. The repository includes the backend services, media service, frontend console, and an all-in-one Docker Compose deployment so users can quickly start a complete visual intelligence platform.

The platform does not bundle an inference engine or depend on a specific model provider. You can configure an OpenAI-compatible vision model API, register cameras or video streams, bind algorithms to devices, and run scheduled analysis tasks from the console.

• Algorithm Management - Manage visual algorithms, prompts, output formats, model bindings, and test settings.
• Model Configuration - Configure OpenAI-compatible model APIs, including endpoint, API key, model id, context length, and maximum output tokens.
• Prompt Templates - Seed and reuse industry-specific prompt templates for common visual inspection scenarios.
• Device Access - Manage device metadata, stream URLs, status checks, and video previews.
• Task Orchestration - Bind tasks, algorithms, devices, and detection regions, then run scheduled frame analysis by sampling interval.
• Media Service - A standalone media service handles stream connections, frame capture, and snapshots.
• Event Records - Alarm results from algorithm execution are converted into searchable event records.
• One-Command Deployment - Docker Compose starts the frontend, Web API, media service, MySQL, Redis, Kafka, and MinIO.

This edition focuses on the core capabilities of a visual algorithm platform:

• visual algorithm management
• model and prompt configuration
• device access and stream preview
• task management with algorithm-device binding
• frame capture and algorithm execution logs
• alarm event records

Scene governance, organization management, and device organization trees are not included in this open-source edition. For commercial edition capabilities or enterprise adoption support, refer to Enterprise Adoption Support.

vision-hub-platform/
├── backend/   # Java / Spring Boot backend services
├── frontend/  # Vue 3 management console
└── docker/    # Docker Compose deployment

Core components:

Docker Compose deployment only requires:

• Docker
• Docker Compose

Running from source requires:

• JDK 17+
• Maven 3.8+ or the included Maven Wrapper
• Node.js 18+
• MySQL 8+
• Redis
• Kafka
• MinIO
• An OpenAI-compatible vision model API

cd docker
cp .env.example .env
# Edit .env to change passwords, ports, and secrets before deployment.
docker compose up -d --build

Open:

http://localhost

Default console account (change the password as soon as possible after first deployment):

Username: unicom
Password: ZJ_Unicom

Default ports:

Useful commands:

docker compose ps
docker compose logs -f vision-hub-web-server
docker compose logs -f vision-hub-media-server
docker compose down
docker compose down -v

docker compose down stops services while keeping data volumes. docker compose down -v removes MySQL, Redis, Kafka, and MinIO data volumes, so use it carefully when resetting an environment.

1. Create the database and import the schema:

   mysql -u root -p -e "CREATE DATABASE vision_hub DEFAULT CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci"
   mysql -u root -p vision_hub < backend/vision-hub-web/vision-hub-web-server/src/main/resources/sql/schema.sql

2. Configure backend environment variables:

   export DB_URL="jdbc:mysql://localhost:3306/vision_hub?useUnicode=true&characterEncoding=utf8&useSSL=false&serverTimezone=Asia/Shanghai&allowPublicKeyRetrieval=true&nullCatalogMeansCurrent=true"
   export DB_USERNAME="root"
   export DB_PASSWORD="ZJ_Unicom"
   export REDIS_HOST="localhost"
   export REDIS_PORT="6379"
   export REDIS_PASSWORD="ZJ_Unicom"
   export KAFKA_SERVERS="localhost:9092"
   export MINIO_ENDPOINT="http://localhost:9000"
   export MINIO_ACCESS_KEY="ZJ_Unicom"
   export MINIO_SECRET_KEY="ZJ_Unicom"
   export MINIO_BUCKET="vision-hub"
   export INTERNAL_SECRET="ZJ_Unicom"
   export JWT_SECRET="ZJ_Unicom"

3. Start backend services:

   cd backend
   ./mvnw -pl vision-hub-web/vision-hub-web-server -am spring-boot:run
   ./mvnw -pl vision-hub-media/vision-hub-media-server -am spring-boot:run

4. Start the frontend console:

   cd frontend
   npm install
   npm run dev

   The local frontend dev server runs at http://localhost:3000 and proxies /api to the Web API service.

The platform calls vision models through OpenAI-compatible chat/completions APIs. Configure the following fields in the console:

• API endpoint, for example http://your-model-server/v1/chat/completions
• API key
• Model id, such as gpt-4o-mini or a model name exposed by your local model server
• Model context length
• Maximum output tokens
• Image transfer mode, URL or Base64 depending on model server capability

Model services, inference GPUs, and third-party API keys are not included in this project. Users need to provide them separately.

• MySQL, Redis, Kafka, and MinIO use Docker named volumes.
• Media snapshots are written to docker/data/media-snapshots by default.
• Uploaded files and algorithm result images are stored in MinIO.
• The bundled frontend nginx proxies /minio/ to MinIO so returned image URLs remain same-origin.

• Do not use default accounts, database passwords, or JWT secrets in production.
• For public deployments, set SWAGGER_ENABLED=false to disable API documentation.
• docker/.env.example is only a local evaluation template. Copy it to .env and change all sensitive values before deployment.

Does Docker Compose deploy all middleware services?

Yes. The Compose stack starts MySQL, Redis, Kafka, MinIO, the frontend, the Web API, and the media service. It is suitable for local evaluation, testing, and small to medium private deployments. Production environments can be adapted to use external middleware.

Does this project include a vision model?

No. The platform handles model configuration, task orchestration, frame capture, and result records. Actual inference is provided by the OpenAI-compatible model service configured by the user.

Should existing databases be upgraded with schema.sql only?

No. schema.sql is mainly for fresh database initialization. Existing deployments should use migration scripts to upgrade table structures or seed new data without overwriting existing data.

Although Vision Hub Platform is open source, real enterprise adoption usually involves more than starting the services. Production projects often require scenario governance, workflow design, algorithm strategy configuration, and integration with existing business systems. This type of work requires professional design based on the on-site business context, network environment, device protocols, data security requirements, and operations requirements.

For enterprise consulting and technical support, please contact us:

• Phone / WeChat: 15657170299 (same number for WeChat)

We can help with end-to-end delivery, including requirement research, scenario governance, workflow solution design, system integration, deployment, and operations support, so visual intelligence capabilities can truly run inside real business workflows.

MIT © ZJ-Unicom-AI