[email protected]
+86 13370882355

3D Vision Sensors Selection Guide

3D Vision Sensors Selection Guide

To help users make informed decisions when choosing among various 3D vision hardware solutions, we have compiled this document for your reference, providing a general analysis of the principles, advantages, disadvantages, and application scenarios of the four mainstream 3D vision sensors: Stereo Vision, Structured Light, iToF, and dToF.

Disclaimer: The descriptions of different technologies in this document are based on their underlying principles and general performance characteristics, and do not represent any specific manufacturer’s product. Different camera manufacturers may optimize their products using different chips, algorithm enhancements, or hybrid approaches, which could result in performance that differs from the general descriptions provided here. Feedback and corrections are always welcome.

I. Stereo Vision (Passive and Active)

Stereo Vision 3D Camera Technology

A passive ranging technology that mimics the human eye’s disparity principle. It is divided into passive stereo and active stereo.

Working Principle

Triangulation-Based Depth Reconstruction

Stereo depth reconstruction uses triangulation to calculate the distance from the object being measured to the camera. When two cameras observe the same object, the positional difference of the object in the two images is called disparity. The closer the object is to the camera, the larger the disparity; the farther away, the smaller the disparity.

Given the known relative positional relationship (e.g., the distance between the two cameras), the distance can be calculated using similar triangle principles.

Passive Stereo vs. Active Stereo Comparison

Aspect Passive Stereo Active Stereo
Principle Relies solely on natural texture from ambient light; calculates disparity through image matching Adds an infrared laser projector (dot matrix/speckle/line) to actively project texture, enhancing matching in low-texture areas
Advantages Low cost, low power consumption, no active light interference, multiple devices can work simultaneously Usable in low-light or textureless scenes; more stable performance than passive stereo
Disadvantages Essentially unusable on low-texture surfaces (white walls) or in low-light conditions Projection patterns interfere with each other when multiple devices work together, causing matching failures; pattern is easily overwhelmed by strong outdoor sunlight, degrading depth performance
Typical Representatives Stereolabs ZED series, Senseno RealSense D400 series, Luxonis OAK series, Orbbec Gemini series
Active stereo is an enhanced version of passive stereo. The infrared fill light solves the pain points of low-light and low-texture conditions. However, it does not fundamentally improve issues such as long-range accuracy or strong light interference. Active stereo still suffers from the inverse-square degradation of accuracy inherent to triangulation methods. Furthermore, when multiple active stereo devices operate simultaneously, projected patterns interfere with each other, leading to matching failures.

II. Structured Light

Structured Light 3D Sensing

An active triangulation technique that actively projects a coded pattern and directly solves for depth through pattern deformation.

Working Principle

Active Triangulation with Coded Patterns

Structured light is an active triangulation measurement technique, belonging to the same broader category as active stereo vision.

The working method is as follows: an infrared laser projector casts a structured light pattern containing coded information (such as stripes, grids, or pseudo-random dot arrays) onto the object being observed. These patterns deform (e.g., stripes bend, dots shift) according to the object’s geometry and distance. An infrared camera captures these deformed patterns, and the system decodes the degree of deformation to directly calculate the depth information for each pixel.

Core Difference from Active Stereo

Aspect Active Stereo Structured Light
Depth Calculation Calculates disparity by matching corresponding points in left and right images, relying on correspondence between images Directly obtains depth by decoding pattern deformation; each projected feature point carries its own “identity tag” and does not rely on stereo matching

Core Advantages

  • Active light source adds texture: Solves the pain points of passive stereo in low-texture and low-light scenes.
  • Extremely high close-range accuracy: Can achieve sub-millimeter accuracy (e.g., Apple Face ID module).
  • No need for stereo matching: Each projected feature point carries coded information, providing a more direct computational path.

Technical Limitations

  • Limited measurement range: Optimal operating distance is typically 0.3-3 meters; long-range accuracy degrades with the square of the distance.
  • Multi-path interference: Multiple reflections in complex scenes can cause depth calculation errors.
  • Interference from similar devices: When multiple structured light devices work simultaneously, their projected patterns interfere with each other, making decoding impossible.
  • Sunlight interference: In strong outdoor light, the projected pattern is easily overwhelmed by ambient light, limiting operation.

Typical Representatives

Consumer Apple Face ID module, Orbbec Astra series (mainly used for close-range face recognition, 3D modeling, etc.)

Industrial Photoneo MotionCam-3D series, Mech-Mind Mech-Eye series, Zivid products

III. iToF (Indirect Time-of-Flight)

iToF Indirect Time-of-Flight Sensor

A ranging technique that indirectly calculates flight time through phase difference.

Working Principle

Phase-Based Distance Measurement

iToF (indirect Time-of-Flight) collects energy values at different time windows, analyzes the proportion between these values, and indirectly measures the time difference between the transmitted and received signals. It is mainly divided into two modulation methods:

CW-iToF (Continuous Wave): Uses sinusoidal wave modulation. The phase shift between the received and transmitted sinusoidal waves is proportional to the object’s distance. Accuracy is limited by random noise and quantization noise. To improve accuracy, high-power, short-integration-time sampling combined with high modulation frequency is typically used.

PL-iToF (Pulse Modulation): Emits light pulses with amplitude and time information. Uses dual-sampling techniques to improve accuracy. The calculation is simpler and has lower computational load, but accuracy is weaker than CW-iToF, and it is more sensitive to background noise.

Core Advantages

  • Mature chip industry chain.
  • Good real-time performance.

Technical Limitations

  • Flying pixels: At object edges, a single pixel simultaneously receives reflected light from both the foreground and background, resulting in incorrect depth values and generating “floating invalid points in the air.”
  • Multi-path interference (MPI): In real-world scenes, complex diffuse or specular reflections cause light to be reflected multiple times before reaching the sensor, leading to systematically overestimated measurements. This has been the biggest technical obstacle plaguing iToF for many years.
  • High sensitivity to black objects: Black objects have strong light absorption and weak reflected signals, causing a sharp drop in signal-to-noise ratio. Depth data easily develops holes or becomes invalid, making it difficult to stably detect dark-colored targets.
  • Severe multi-device interference: When multiple iToF devices operate simultaneously in the same area, their modulated signals interfere with each other, causing large-scale errors or complete failure of depth data, limiting deployment in dense cluster scenarios.

Typical Representatives

iToF chip manufacturers: Sony, Infineon, PMD.

Module manufacturers: Lucid Helios2 series, SICK Visionary-T Mini, MRDVS M series, IFM 3D cameras.

IV. dToF (Direct Time-of-Flight)

dToF Direct Time-of-Flight LiDAR

The ultimate ranging technique that directly measures the round-trip time of photons.

Working Principle

Single-Photon Detection with TCSPC

dToF (direct Time-of-Flight) technology directly measures the time difference between emitting and receiving a light pulse. Due to laser safety limits and consumer product power constraints, the pulse energy emitted by ToF cameras is limited. By the time the light returns to the receiver, the energy density has decreased by more than a trillion times. Ambient light acts as noise, severely interfering with signal detection.

Therefore, dToF requires extremely sensitive photodetectors: Single-Photon Avalanche Diodes (SPADs). When a SPAD is in operation, it is biased with a high reverse voltage. When a photon is absorbed and converted into a free electron, the strong internal electric field accelerates this electron, which collides to generate more carriers, creating a geometrically amplified avalanche effect, thereby outputting a large current pulse and achieving single-photon detection.

dToF uses the Time-Correlated Single-Photon Counting (TCSPC) method to achieve picosecond-level time precision. The system repeats the emission-detection of the same pulse signal thousands to hundreds of thousands of times, obtains a statistical distribution histogram from each detection, reconstructs the curve of light pulse energy over time, and thus derives the precise flight time.

Core Advantages

  • Distance accuracy: Within the normal operating range, error does not significantly amplify with increasing distance, maintaining centimeter-level accuracy at medium-to-long ranges.
  • Multi-path interference suppression: Directly measures the arrival time of the first wave, making it much less affected by multi-path reflections than iToF.
  • Low power consumption and high efficiency: A single pulse completes the distance measurement, with extremely low computational load and lower latency.
  • Ambient light adaptability: Uses time-gating technology combined with narrow-band filters and near-infrared light sources (typical wavelength 940nm), enabling effective depth data acquisition under high ambient light conditions up to 100kLux.
  • Multi-device collaboration friendly: Pulse timing of different devices can be staggered (e.g., using time-division or pulse coding), natively supporting simultaneous operation of multiple devices.
  • Real-time performance: No need for complex matching algorithms; depth data can be output directly, with typical frame rates of 10-20 fps.

Technical Challenges (Gradually Improving)

  • SPAD Dark Count Rate (DCR): Improved compared to early solutions through 3D stacking processes and quenching circuit optimization.
  • Photon Detection Efficiency (PDE): The application of BSI (backside illumination) technology has improved photosensitivity.
  • On-chip integration: With advances in advanced process nodes, SPAD arrays, TDCs (Time-to-Digital Converters), and histogram algorithms can now be integrated on-chip.
  • Relatively low spatial resolution: Limited by the physical size of SPADs and the complexity of backend counting circuits (TDCs), resolution is generally lower than iToF.

Typical Representatives

Consumer/Mobile Apple LiDAR (iPad Pro/iPhone Pro series).

Industrial RoboSense E1/E1R, MRDVS S series.

Share to:
Recently Released
Contact Us