Heard some friends have been "misunderstood" by electronic surveillance while driving.
Some wearing black clothes were mistakenly judged as not wearing seat belts.
Some scratching their ears were mistakenly identified as making phone calls.
Some mobile phones placed on a stand were mistakenly identified as being used for playing games.
……
It's time to test your eyes!↓
Misjudgments in these complex scenarios are a headache for drivers.
It's also a challenge for traffic violation detection services.
Bringing in more人工 review volume
May also affect the accuracy of penalty decisions.
To enhance the accuracy of traffic checkpoint photo detection
HikvisionBased on the HiKvision ViewLynx large model technology framework
Deploy large model capabilities directly to
Traffic Intersection Capture Series
Compared to traditional detection algorithms
Seat belts, making phone calls, using mobile phones
Error detection rate reduced by over 75%.
(As per actual project data)
Enhanced with a visual large-scale model
Traffic Intersection Capture Product Line
Eliminate "misunderstandings" more effectively at the machine detection stage
Ease the burden of manual review
Better identify those not wearing seat belts,
Distracted driving, such as using a phone while driving
Boosting Traffic Safety Management Efficiency
Check out the comparison below ↓↓↓
To be sure about it
These images are so blurred that they're almost unrecognizable to the naked eye.
How are large models distinguished?
How to Further Improve Accuracy
From the parts to the whole
Traffic checkpoint camera systems become more "sensitive"
In traffic violation detection, traditional deep learning algorithms break images into pieces, much like solving a puzzle, examining local details first before piecing together the whole. When identifying seatbelt usage, false positives are common due to low contrast, obstructions, and complex postures. Similarly, when detecting phone usage, false alarms occur easily due to raised hands, holding objects, or non-straight gazes.
Enhanced with a visual large-scale model
Safety Belt Inspection:Through large model global association and semantic understanding, even when seat belts are obscured or there's a misconception caused by non-belt elements like wipers, accurate identification can be achieved by a comprehensive judgment of human posture, remaining visible parts, contours, and the human figure. With massive pre-trained data and deep-level structures, the system also demonstrates enhanced adaptability to complex scenarios and non-standard belt shapes, such as precise recognition of belt covers.
Phone Usage Monitoring:By leveraging the self-attention mechanism of large models, we no longer over-rely on local features, such as phone shape, and instead analyze human-body contact, gaze, interaction actions, and vehicle structure. This accurately identifies the correlation between the phone and the driver.
Next, let's put our eyes to the test against the large model.
Determine if the driver in the image is wearing a seatbelt, making a phone call, or using a mobile device.
Round 1 Seatbelt Inspection
Round 2 phone call verification
Round 3 Mobile Phone Detection
Traffic Intersection Capture Camera Product Matrix
The Hikvision Traffic Intersection Capture Series, powered by the visual large model, enhances the full perception capability of complex traffic scenarios by deeply挖掘 the potential relationships between different pieces of information. It overcomes performance bottlenecks brought by low contrast, obstructions, and complex poses in complex environments, meeting diverse traffic management needs.
Not Just Toll Booth Capture Series Products
Hikvision leverages the Guanlan large model technology system
We will continue to innovate and consistently develop.
Enhancing Smart, Safe, and Efficient Transportation Scenarios
"Many roads lead, but safety is the only one."
Always wear your seatbelt when driving.
Don't use your phone or make calls
Safe Driving, Safe Journey
