Dahua Technology introduces Xinghan Large-Scale AI Models for constant evolution towards an intelligent AIoT future

Dahua Technology, a global leader in video-based AIoT solutions and services, has officially introduced its Xinghan Large-Scale AI Models — a next-generation industrial-grade AI system that combines large-scale visual intelligence with multimodal and linguistic capabilities. Xinghan was designed to meet the complex challenges of real-world environments, representing a significant advancement in Dahua’s continuous innovation efforts that enable intelligent transformation across various industries.

The Technological Foundation of Xinghan

With the mission of enabling machines to truly understand the world, the evolving Xinghan model system serves as a bridge between cutting-edge research and practical applications. Named after the Chinese word for „galaxy,“ Xinghan offers a full-stack capability matrix based on Edge-Cloud synergies, enabling scalable adaptive intelligence across different industries. The updated Xinghan architecture comprises three core model series: L, V, and M. The L-series model focuses on natural language understanding and interaction while the other two models cover more specific applications:

V-Series: Xinghan Vision Models

This series emphasizes advanced visual intelligence and video analysis by optimizing target categories such as people, vehicles, and non-vehicles to reduce model complexity while ensuring high accuracy.

• Perimeter Protection: Expands coverage by precisely identifying smaller targets (even down to 20×20 pixels) compared to traditional CNN-based AI models, reducing false alarms and increasing detection range of large-model cameras.*
• WizTracking: Provides an intelligent next-generation tracking algorithm capable of handling complex occlusions and variations in target position with 50% improved accuracy.*
• Crowd Map: Significantly enhances recognition of small targets at long distances (up to 2x further), features umbrella compensation increasing accuracy by 80% during rain conditions. It also offers a 2.5x enlargement of analysis area supporting detection up to 5,000 individuals with robust performance in crowded low-light environments.*
• Scene Adaptive – AI WDR: Utilizes situational awareness for analyzing both spatial and contextual scene features allowing intelligent automated camera configuration.
• AI Rule Assist: Developed for automatic rule delineation against intruders providing one-click access high-precision scene recognition automatic analysis among others.

M-Series: Multimodal Xinghan Models

Multimodal models are advanced AI systems capable of processing multiple heterogeneous data types simultaneously (e.g., text images audio video) deeply integrating them improving information processing efficiency enabling more natural human-computer interaction opening up broader application scenarios.

• WizSeek: Revolutionizes video search through natural language queries simply describe your target (e.g., people vehicles animals objects etc.) WizSeek instantly finds relevant footage within recorded video archives.
  

• Text -Defined Alarms:Users can define alarms by describing them naturally lowering development thresholds significantly allowing quick flexible scalable configurations tailored different real-life scenarios.