A library of 50+ ready-to-deploy video analytics algorithms and on-premise AI large models, pre-adapted for TouchVision SOPHGO edge AI boxes and servers. It turns existing IP cameras into safety and operations sensors without a cloud subscription.
Overview
TouchVision’s algorithm library gives every YZ-series edge AI box and AI server a head start: more than 50 video analytics algorithms are ready to run on existing IP cameras, with no cloud subscription and no video leaving the site. Algorithms cover personal protective equipment, fire safety, security, staff behaviour, traffic and facility management, and they can be combined per camera and schedule.
Alongside vision, the library includes on-premise AI large models: DeepSeek and Qwen for chat and knowledge Q&A, Pangu large models, SDXL text-to-image, CLIP text-to-image and text-to-video retrieval, CodeGeeX2 code generation, Whisper speech-to-text and SadTalker digital humans. Events are pushed to your platform over HTTP REST or MQTT, and custom algorithms can be trained with the AI Model Training & Inference Platform.
Key features
- 50+ video analytics algorithms, adaptable to each site
- PPE: safety helmet, high-visibility vest, non-standard workwear
- Fire safety: smoke, open flame, smoking
- Security: intrusion, loitering, fighting, person fall, face recognition
- Staff behaviour: off-post, sleeping on duty, phone calls, smartphone use
- Traffic and facilities: illegal parking, vehicle congestion, wrong-way driving, passage congestion, e-bike in elevator, road waterlogging, water level, street garbage, rodents, warning signs
- Large models: DeepSeek, Qwen, Pangu, SDXL, CLIP, CodeGeeX2, Whisper, SadTalker
- Custom algorithms and SDK interfaces for secondary development
Specifications
| Item | Specification |
|---|---|
| Algorithms | 50+ video analytics algorithms |
| Hardware | TouchVision YZ-series edge AI boxes, AI edge servers, 2U AI servers, AI modules |
| Frameworks | Caffe, TensorFlow, PyTorch, MXNet, PaddlePaddle, ONNX |
| Large Models | DeepSeek, Qwen, Pangu, SDXL, CLIP, CodeGeeX2, Whisper, SadTalker |
| Integration | HTTP REST API, MQTT, NVR module, alarm management, voice broadcast |
| Customization | Custom algorithms and SDK |
Specifications are based on manufacturer data and may change. Ask for the latest datasheet for your configuration.
Applications
- Smart gas stations
- Industrial parks and campuses
- Residential communities
- Hospitals and business halls
- Power grid sites
- Commercial kitchens
- Construction sites
- Chemical plants
Related industry solutions: AI Video Analytics & Smart Safety · Fuel Dispensers & Stations · Fire Safety & Building Control
Customization and ordering
The AI algorithm library can be customized for OEM projects, and small-batch pilot orders are supported. Request a quote or sample, chat on WhatsApp (+86 135 0532 6001), or read about our OEM/ODM service.
More in this category: Edge AI Computers & Video Analytics Boxes · Related: On-Premise LLM & AI Agent Appliances
Frequently asked questions
Which algorithms are included?
More than 50, including safety helmet, high-visibility vest, smoke, open flame, intrusion, person fall, fighting, loitering, off-post, sleeping on duty, phone use, illegal parking, wrong-way driving, e-bike in elevator and water-level detection.
Does the analysis run in the cloud?
No. Algorithms run locally on TouchVision edge AI boxes and servers, and only events and alarms are forwarded over HTTP or MQTT.
Which hardware runs the library?
The YZ-series edge AI boxes, AI edge servers, the 2U AI Training & Inference Server and the AI Edge Computing Module.
Can TouchVision develop a custom algorithm?
Yes. Custom algorithms can be trained with the AI Model Training & Inference Platform and delivered with SDK interfaces.