AI-Powered Vehicle Classification & Traffic Intelligence on the Malaysia North-South Expressway
- Metadyne Admin
- Nov 25
- 2 min read
Intelligent Vision for the North–South Expressway
MetaDyne has engineered and successfully developed a fully in-house AI Vision Vehicle Classification and Counting System on Malaysia’s PLUS North–South Expressway, purpose-built for highway-scale, real-time traffic intelligence. The system transforms high-resolution video feeds from existing IP cameras into structured traffic data—accurately detecting, classifying, and tracking vehicles at highway speeds across multi-lane roadways. Importantly, the system is proven to perform in challenging nighttime environments, where conventional detection methods often fail.
Class-Accurate Vehicle Identification Powered by Deep Learning
The heart of the system is MetaDyne’s custom-trained convolutional neural network (CNN), designed not only to detect vehicles but to classify them into Malaysia’s recognized tolling classes based on their physical geometry, motion profile, and axle configuration.
The system distinguishes among:
Class 1: Cars and passenger vehicles (2 axles, 3–4 wheels)
Class 2: Light lorries (2 axles, 5–6 wheels)
Class 3: Heavy lorries (3 or more axles)
Class 4: Taxis (car body, yellow plate)
Class 5: Buses (multi-axle, extended height and length)
Class 6: Motorcycles (narrow profile, rapid lane shifts)
Using deep visual features such as wheelbase length, shadow segmentation, height mapping and motion vectors, the AI Vision model goes beyond bounding box estimation and silhouette shape. This results in high-confidence classification, even during fast motion, vehicle clustering, or partial occlusion.
Exceptional Performance in Nighttime Conditions
A key strength of MetaDyne’s platform is its ability to operate reliably under low-light and nighttime conditions, which are often the most challenging for vision-based systems.
The AI engine has been tuned using night-specific datasets, including:
Headlight glare normalization
Adaptive exposure control
Temporal smoothing under strobe or IR lighting
Shadow-resilient vehicle edge detection
The system can track and classify vehicles even when headlights saturate the image or when ambient lighting is minimal. This capability significantly outperforms traditional sensor methods (like loop detectors or beam counters), which offer no visual classification and are blind to axle count and shape at night.
End-to-End Vehicle Tracking with Virtual Count-Lines
Once detected, each vehicle is assigned a unique identifier and tracked across the frame using motion prediction algorithms and IoU-based re-identification. MetaDyne’s virtual “count lines” record the exact moment the vehicle crosses a designated boundary, logging the class, direction, timestamp and lane ID. This enables structured, high-frequency traffic data aggregation without the need for any road-side embedded hardware.
Scalable Deployment Without Road Modification
One of MetaDyne’s key innovations lies in its non-intrusive deployment model. All analytics are performed on video streams from standard roadside IP cameras—no ground-loop sensors, lidar gates, or physical reworks are required. Whether installed on gantries or poles, the cameras can be calibrated for multi-lane operation using a single configuration pass.
This makes the system ideal for deployment along high-traffic corridors like the North–South Expressway, where road closures for sensor installation would be cost-prohibitive or disruptive.
A Platform for Intelligent Transport Evolution
With MetaDyne’s fully in-house AI Vision architecture, future enhancements such as lane-drift detection, stopped vehicle alerts, wrong-way driver detection and speed-based violation detection are already under prototyping.
By turning cameras into intelligent sensing platforms, MetaDyne enables Malaysia to move toward a fully AI-integrated expressway ecosystem, optimized for safety, efficiency and long-term infrastructure planning.
Comments