 |
HANGZHOU, China, Sept. 19, 2025 /PRNewswire/ — Dahua Technology, a world-leading video-centric AIoT solution and service provider, has officially launched its Xinghan Large-scale AI Models — a next-generation, industry-grade AI system that integrates large-scale visual intelligence with multimodal and language capabilities. Developed to address the complex challenges of real-world environments, Xinghan represents a major leap forward in Dahua’s continuous innovation, empowering intelligent transformation across diverse sectors.
The Dahua Xinghan Large-Scale AI Model places visual analysis at its core, integrating multimodal capabilities and embedding deep industry expertise to create a large model tailored for diverse application scenarios. Built upon this real-world foundation, it serves as the key to achieving scalable and commercially viable AI solutions.
Technological Foundation of Xinghan
With the mission of enabling machines to truly understand the world, the Xinghan model system continues to evolve by bridging cutting-edge research with real-world applications. Named after the Chinese word for “galaxy”, Xinghan delivers a full-stack capability matrix powered by edge-cloud synergy, enabling scalable, adaptive intelligence across industries. The upgraded 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 tackle more specific applications:
V-Series: Xinghan Vision Models
Centered on advanced visual intelligence and video analytics, this series streamlines target categories by focusing on key targets (e.g. humans, motor vehicles, and non-motor vehicles) to reduce model complexity while maintaining high accuracy.
Key features include:
- Perimeter Protection: The coverage is extended by accurately identifying smaller targets (even down to 20×20 pixels) compared to traditional CNN-based AI models, reducing false alarms and increasing the detection range of large-model cameras.*
- WizTracking: It offers a next-generation intelligent tracking algorithm that can handle complex occlusions and variations in target posture, achieving 50% improvement in accuracy.*
- Crowd Map: It significantly enhances small-target detection at long distances (up to 2× farther) and features umbrella compensation, improving accuracy by 80% during rainy conditions*. It also offers a 2.5× increase in analysis range, supports detection of up to 5,000 people, and provides robust performance in dense crowds and low-light environments.*
- Scene Adaptive – AI WDR: It leverages situational awareness to analyze both spatial and contextual characteristics of a scene, enabling intelligent and automated camera configuration.
- AI Rule Assist: It is designed for the automatic delineation of Perimeter Protection intrusion rules, offering one-click access, highly accurate scene recognition, automatic analysis, and more.
M-Series: Xinghan Multimodal Models
Multimodal models are advanced AI systems capable of simultaneously processing and deeply integrating multiple heterogeneous data types (e.g. text, images, audio, and video). This significantly enhances the efficiency of information processing, enables more natural human-computer interaction and unlocks a broader spectrum of application scenarios.
Key features include:
- WizSeek: It revolutionizes video investigation through natural language search. Simply describe your target (e.g. people, vehicle, animal or item, etc.) and WizSeek instantly retrieves matching footages across recorded video archives.
- Text-Defined Alarms: It allows users to define alarms by simply describing them in natural language, which significantly lowers the development threshold, and enables fast, flexible, and scalable configuration tailored to diverse real-world scenarios.
For more information about the Xinghan Large-scale AI Models, please visit here.
*Results are based on standard setup and testing environment.
