Stereo Vision Depth Mapping for People Counting Sensors
Introduction: Stereo depth mapping lets a dual-lens people counter measure distance, so shadows, carts, and luggage have less power to distort counted foot traffic.
A low-voltage technician setting up a visitor counter is giving the system a way to see space. A 2D camera reads shapes and brightness, so a dark floor patch can look like a person. A stereo sensor compares two views and assigns distance to scene points. Disparity and calibration build the depth map that keeps footfall counting stable at real entrances.
Why a Pair of Lenses Creates Depth Information
A single lens captures a flat projection. It can recognize a person after training, but each pixel lacks direct distance. Shadows, reflections, and floor patterns can confuse a 2D model because they change brightness without changing physical depth. Two lenses change the problem by viewing the same doorway from slightly different positions, much like human eyes. The small horizontal offset creates a measurable shift for every visible object. That shift is disparity. Nearby objects produce a larger shift between left and right images; distant objects produce a smaller shift. The sensor finds matching points and converts pixel shift into distance. This is the core of stereo depth mapping: two viewpoints, one scene, and a known distance between lenses. The CARDLAN CL-W6 applies this architecture in a commercial people counter with dual lenses, a Sony image sensor, and an onboard AI chip rated at least 1.0 Tops. Its mounting range is 1.8 m to 10 m, with coverage from 0.5 m to 15 m. Those figures matter because stereo depth works best when the entrance layout gives both lenses a clear view of the walking surface. The device counts anonymous traffic only, so depth data supports counting without storing identity information. Technicians watch the depth map as people, carts, and shadows move through the scene. A shadow lacks a stable stereo match and sits outside the human depth band. A cart has physical shape, yet its depth profile is lower and flatter than a walking person. Depth separates real foot traffic from visual noise by locating each object in 3D space.
How Calibration Keeps Stereo Matching Stable
Calibration makes two lenses behave like one measuring instrument. Even factory-matched lenses differ slightly in position, angle, and distortion. Calibration measures the real relationship between left and right cameras and stores values the processor uses for matching. OpenCV's camera calibration documentation describes this process for stereo systems: it estimates intrinsic parameters for each camera and the geometric relationship between them.
1. How Rectification Aligns the Two Views Before Stereo Matching
Rectification uses calibration values to transform both images so matching points appear on the same horizontal line. This step reduces search complexity. Instead of scanning the whole image, the processor scans along a known line. In a people counting sensor, rectification makes fast, stable stereo matching practical on an embedded AI chip. Poor rectification near shiny floors or glass doors creates depth flicker, holes, or broken outlines; strong rectification keeps the depth surface coherent as people move.
2. How Calibration Values Convert Pixel Shift into Real Depth
After rectification, the system measures disparity in pixels. Calibration converts that pixel shift into a real-world depth estimate using the baseline between lenses and the focal length. This is the mathematical bridge from apparent separation to physical distance. In footfall counting, consistency matters more than raw precision. A stable depth band for people lets the counter apply rules such as counting objects that enter the walking zone between 0.5 m and 2 m from the floor. A drifting calibration can shift that band, causing missed counts or double counts at busy times. Calibration remains part of the measurement chain after installation, because vibration, temperature change, and accidental knocks can move a lens slightly. A commercial counter should hold calibration well, while installers still mount it securely and avoid aiming it at surfaces that confuse stereo matching, such as mirrored walls or direct sunlight. Depth quality depends on setup and environment, so clean installation is part of the measurement chain.
How Depth Data Helps Separate People from Carts and Shadows
A depth map gives the counting algorithm a 3D scene instead of a 2D picture. That changes how the system treats common entrance clutter. A floor shadow lacks a stable 3D body; it may be dark, yet it produces no consistent disparity match at human height. A shopping cart has depth, but its shape is wide, low, and often made of thin bars that create sparse stereo matches. Luggage has a compact, low profile and moves close to the floor. A person has a tall vertical mass with a head and shoulders that usually sit in a different depth band. Installers often describe the difference in practical terms: when the depth map is working well, people appear as distinct upright shapes, while carts and bags appear as smaller or flatter blobs. The counter can then apply depth and size thresholds without recognizing a brand of luggage or exact model of cart. It compares the moving object with the expected 3D profile of a human walking through the entrance. This is why a dual-lens people counter can reduce miscounts from shadows, carts, and luggage while relying on anonymous data. Depth also helps with occlusion, a major source of error in busy doorways. In a 2D view, two people walking side by side can merge into one wide shape. A stereo sensor sees that one person is slightly closer and the other slightly farther, so the depth map can separate them as two bodies at two distances. Similar logic helps when a person carries a large box: box and person may share a 2D outline, while their depth layers differ. The counting model can focus on the human-shaped layer and ignore the carried object. Every entrance has its own lighting and layout, so installers verify the depth map on site. Depth gives the algorithm a stronger signal than brightness alone. The practical value for a low-voltage technician is easier troubleshooting. Unstable counts often trace to the depth chain: a shifted mounting angle, a dirty lens, expired calibration, or a reflective surface can change the stereo match before the AI counts. A clean depth map gives the AI a stable world to count. That is the real difference between disparity-based depth measurement and ordinary 2D recognition: one measures space, the other guesses from appearance.
Conclusion
Stereo vision depth mapping turns a people counting camera into a measuring device. Two lenses create disparity, calibration turns disparity into reliable depth, and the depth map helps the counter separate people from shadows, carts, and luggage. For technicians, calibration and installation quality are part of the counting system rather than background details. A dual-lens sensor such as the CARDLAN CL-W6 uses this approach with a Sony image sensor and onboard AI compute while counting anonymous traffic only. Understanding the depth chain makes it easier to explain why a stereo counter behaves differently from a 2D camera and how to keep its data stable in a real entrance.
FAQ
Q：What is stereo disparity in a people counting camera?
A：Stereo disparity is the horizontal pixel shift between the left and right images for the same point in the scene. Nearby objects shift more; distant objects shift less. A people counting camera uses that shift to calculate depth, so it can tell where a person, cart, or shadow sits in 3D space instead of relying only on a flat 2D image.
Q：Why does camera calibration matter for footfall counting?
A：Calibration defines the true geometric relationship between the two lenses. It supports rectification, which lines up the images for matching, and converts pixel disparity into a depth estimate. Correct calibration keeps the depth map stable as people walk through the entrance; drifting calibration can cause missed counts, double counts, or reactions to shadows and reflections.
Q：Can stereo depth mapping reduce miscounts from shopping carts and luggage?
A：Yes, stereo depth mapping gives the counter a 3D profile to work with. Carts and luggage tend to create low, flat, or sparse depth shapes, while people create taller upright shapes. The algorithm can use those depth differences to filter out non-human objects. Performance still depends on installation, lighting, and calibration, but depth adds a strong layer of separation that a 2D camera does not have.
Sources / References
OpenCV: Camera Calibration and 3D Reconstruction
OpenCV: Depth Map from Stereo Images
Comments
Post a Comment