Product testing: Hikvision’s intelligent checkpoint cameras powered by ‘Guanlan’ AI

Hikvision has recently launched a series of intelligent checkpoint cameras powered by the company’s large model AI, known as “Guanlan.” Designed for traffic applications, the cameras reduce false detection rates and add intelligence to urban traffic management. This article takes a closer look at the cameras, based on product testing by a&s China.

Bringing large model AI directly to the road

 
The cameras include the iDS-TCVC00-HEM model and the iDS-TCVC00-HEL model, both bringing large model AI directly to the road. Compared with traditional algorithms, AI trained on large datasets does a better job identifying dangerous driving behavior such as using smartphones while driving or not wearing a seatbelt, thus enhancing traffic safety management.
 
During nighttime, the iDS-TCVC00-HEL model uses LED strobe lights to ensure capture clarity. The more advanced model, iDS-TCVC00-HEM, adopts multi-spectral fusion technology and can still obtain full-color images when using infrared fill light at night, solving the problem of white light pollution.
 
As for other hardware specs, each model operates in a -30℃-70℃ temperature range and can withstand harsh environments with dustproof and waterproof features. In terms of imaging capabilities, the cameras use high-frame rate global exposure CMOS sensors to achieve dynamic capture without blurring, and license plate character details can still be retained in strong backlit scenes. In terms of video encoding capabilities, the cameras support H.265, H.264 and MJPEG, achieving low latency, low bit rate and high compression capabilities.

Large model AI empowers accurate image recognition

 
Guanlan AI is the soul of the cameras and can identify details such as a vehicle’s make, model and color. Yet this is small cake compared to other things the AI can do.
 
In actual testing, the most amazing thing was the accuracy with which certain driving behaviors were identified. First, seatbelt detection was performed on a driver wearing a black jacket with the seatbelt on. Traditional algorithms inaccurately gave a “seatbelt not on” alert due to low contrast. The Guanlan AI, on the other hand, made the correct judgment by analyzing the sitting position of the human body, the contour of the seat belt and the buckle position. Overall, the large-model AI reduced false detection rates by 75 percent compared with traditional algorithms.
 
The second test had to do with determining whether a driver was talking on the phone while driving. The AI takes into consideration hand movements, sight direction and in-car structure to ensure accurate detection. Even if the mobile phone is blocked by the steering wheel or the picture is blurred, the AI can still make the correct judgment using clues such as charging cables and headphone cables. In the actual test, the AI successfully distinguished between the driver talking on the phone and the driver looking at the dashboard, effectively reducing false alarms.

Traffic violations detection

 
Further, the AI-enabled cameras can accurately identify traffic violations such as illegal lane crossing, illegal lane changing, not wearing a seatbelt, driving at low speed, cutting in, occupying motor vehicle lanes, and making phone calls while driving, among others. In actual tests, the detection of these behaviors was responsive and accurate. It should also be noted the cameras can integrate with radar, vehicle detection processors and other supplementary systems to meet the user’s unique and specific needs.

Applications in different scenarios

 
The cameras adopt an energy-saving design and support local TF card backup storage (256GB), which ensures the integrity of data in the event of network disconnection. The cameras can be applied in a variety of scenarios. For example, installed on a city’s main roads, the cameras can enable smart policing by capturing illegal behaviors such as speeding, crossing the line, driving in the wrong direction, and drivers not wearing a seatbelt. When deployed near schools, the cameras can strengthen student safety by detecting drivers driving and talking on the phone at the same time.

Product features
 

• Hardware: Resists dust and water, protects against lightning and surge, works in a wide temperature and voltage range;  
• Night vision: Captures clear images via dual infrared and visible light channels, improves feature extraction in dark environments by 40 percent;  
• Imaging: Leverages high-frame rate global exposure CMOS sensor to achieve dynamic capture without ghosting, retains license plate character details in backlit scenes;  
• Ease of deployment: Deploys easily as capture, identification, storage and uploading are performed by one device;  
• Intelligent analysis: Realizes extraction of fully structured data such as motor vehicles, non-motor vehicles and other targets;  
• Algorithms: Breaks away from traditional local feature extraction and achieves accurate driving behavior detection through semantic understanding.  

a&s comments

 
As the world's leading AIoT solutions provider, Hikvision has achieved another milestone by launching the Guanlan AI intelligent checkpoint cameras for traffic applications. Whether used on a city throughway or rural road, regardless of night or day, fog or rain, the cameras can achieve accurate detection, reduce manual review and improve law enforcement efficiency. Indeed, as smart traffic management evolves from "visible" to "clear" to "understandable,” the cameras stand ready to provide an ideal solution that makes roads safer for all of us.