Phase Shifting Method, Structured Light Systems, and Time of Flight Method

While discrete binary patterns allow unambiguous triangulation, achieving sub-pixel 3D accuracy requires projecting continuous intensity functions across the scene. In this chapter, we explore continuous phase shifting, high-profile industrial structured light applications, fundamental optical limits, and Time-of-Flight (ToF) range sensing.


1. Phase Shifting Method

Rather than projecting discrete binary stripes, the phase shifting method projects mathematical light patterns whose intensities vary continuously across space. This increases spatial resolution to sub-pixel accuracy.

1.1 Intensity Ratio Method

  • Ramp Function: A single ramp illumination pattern $L_1$ is projected onto the scene, where intensity decreases linearly from maximum brightness at one side to zero at the other ($x_p$).
  • Flat Uniform Illumination: A second image is captured under flat uniform light $L_2$.
Intensity Ratio Method Projection Patterns
Figure 1: Projection of linear ramp pattern L1 and flat uniform pattern L2 [Carrihill 1985].
  • Normalization: Measuring pixel intensities $I_1 = \rho \cdot L_1$ and $I_2 = \rho \cdot L_2$ in the camera and taking their ratio cancels the unknown surface albedo and surface normal factor $\rho$:

$$\frac{I_1}{I_2} = \frac{\rho \cdot L_1}{\rho \cdot L_2} = \frac{L_1}{L_2}$$

Intensity Ratio Normalization
Figure 2: Ratioing I1/I2 eliminates albedo variation, yielding direct mapping to projector column coordinate x_p.
  • Disadvantage: Highly sensitive to sensor noise and projector intensity quantization steps.

1.2 Sinusoidal Phase Shifting Mathematics

In industrial automation and quality inspection, the gold-standard technique is Sinusoidal Phase Shifting, which projects continuous cosine waves onto the scene and shifts their phase temporally.

The emitted projector cosine wave is defined by average brightness $b$, amplitude $b$, and period $P$. Accounting for unknown ambient lighting $a$ and relative surface albedo $\rho$, the pixel intensity observed by the camera is:

$$I_1(x_c, y_c) = \rho a + \rho b + \rho b \cos\left( \frac{2\pi x_p}{P} \right)$$

Sinusoidal Cosine Wave Projection
Figure 3: First reference cosine wave L1 projected onto the scene [Wust 1991].

This equation contains three unknowns: $\rho a$ (ambient component), $\rho b$ (amplitude component), and the target projector column coordinate $x_p$. To solve for these three unknowns, exactly three phase-shifted images are captured:

  1. Frame 1 ($I_1$): Reference cosine pattern $L_1$ projected with $0^\circ$ phase shift.
  2. Frame 2 ($I_2$): Pattern phase shifted by $-120^\circ$ ($-2\pi/3$).
Phase Shift -120 degrees
Figure 4: Second cosine pattern L2 shifted by -120° (-2π/3).
  1. Frame 3 ($I_3$): Pattern phase shifted by $+120^\circ$ ($+2\pi/3$).
Phase Shift +120 degrees
Figure 5: Third cosine pattern L3 shifted by +120° (+2π/3).

Solving these three simultaneous trigonometric equations eliminates ambient lighting $\rho a$ and amplitude $\rho b$, yielding a closed-form solution for projector column $x_p$:

$$x_p = \frac{P}{2\pi} \tan^{-1}\left( \sqrt{3} \frac{I_2 - I_3}{2I_1 - I_2 - I_3} \right)$$

Phase Shifting Closed-Form Solution
Figure 6: Closed-form trigonometric solution for projector column coordinate x_p using 3 phase-shifted images.

Intersecting the computed projector column plane $x_p$ with the camera viewing ray yields sub-millimeter 3D point accuracy.


2. Structured Light Systems

2.1 Notable High-Profile Systems

  • 3D Visual Inspection (Omron Corp.): Used in surface-mount factory assembly lines to inspect printed circuit board (PCB) solder joints and micro-components in real time. The PCB is tiled and scanned via phase shifting in seconds to reject defective solder joints instantly.
  • Digital Michelangelo Project (Levoy 2000): Stanford researchers scanned Michelangelo’s David statue in Florence over 30 nights using precision structured light range scanners. Achieving a mesh resolution of $1/4 \text{ mm}$, the project created a permanent digital twin (Virtual David) for micro-erosion tracking and archival preservation.
Digital Michelangelo Project Scanning David Statue
Figure 7: Digital Michelangelo Project: High-resolution 3D mesh (1/4 mm accuracy) of Michelangelo's David statue [Levoy 2000].
  • Great Buddha Project (Ikeuchi 2007): Drone-mounted structured light and laser scanners were deployed in Nara, Japan to digitize the monumental Great Buddha statue and surrounding temple heritage structures.
Great Buddha Project
Figure 8: Great Buddha Project: Physical statue in Nara and its 3D digital model [Ikeuchi 2007].

2.2 Limitations and Unsolved Problems

Despite high accuracy, structured light systems face physical limitations on certain surface and material types:

  1. Specular / Metallic Surfaces: Mirror-like specular reflection redirects light exclusively along the angle of reflection. Light rarely backscatters to the camera, leaving empty holes in the depth map.
  2. Translucent / Subsurface Scattering Surfaces: On materials like marble, wax, or human skin, light penetrates beneath the surface and scatters internally before exiting from adjacent pixels, destroying pattern edge sharpness.
  3. Participating Media: In fog, smoke, or turbid underwater environments, light attenuates rapidly and ambient scattering causes the medium itself to glow, masking projected patterns.
  4. Transparent Objects (Glass/Water): Light refracts directly through glass objects without scattering.
  5. Hair and Micro-Fibers: Hair strands are far smaller than an individual camera pixel, causing multiple strands to project onto a single pixel and breaking geometric triangulation.
Unsolved Problems in Structured Light
Figure 9: Challenging surfaces for structured light: Subsurface scattering (marble), participating media (underwater), specular metal, transparent glass, and hair.

2.3 Summary of Structured Light Methods

The table below summarizes the image count complexity of all major structured light range finding paradigms:

Structured Light Methods Summary Table
Figure 10: Summary table comparing required image frame counts across structured light paradigms.

3. Time of Flight Method (ToF)

Time of Flight (ToF) range sensing bypasses baseline triangulation entirely by directly measuring the round-trip travel time of light ($c \approx 3 \times 10^8 \text{ m/s}$).

3.1 Biological Origins and Historical Speed of Light Experiments

  • Biological Biosonar: Bats, dolphins, and whales use echolocation (sonar) by emitting sound waves and timing returning echoes to perceive 3D space. ToF applies this exact principle using light.
Echolocation in Nature
Figure 11: Biological origins of Time-of-Flight: Echolocation using sound waves in bats, dolphins, and submarines.
  • Galileo’s Lantern Experiment (1600s): Galileo attempted to measure light speed by placing two lantern operators on hilltops 1000 meters apart (2000m round trip). Since light travels 2000m in just $6.6 \ \mu\text{s}$, human muscle reflexes (~milliseconds) rendered the experiment unsuccessful.
Galileo's Speed of Light Experiment
Figure 12: Galileo's early attempt to measure the speed of light across 1000m hilltop baselines.
  • Fizeau’s Cogwheel Experiment (1849): Hippolyte Fizeau successfully measured light speed by passing light through a rapidly spinning cogwheel over an $8633 \text{ m}$ distance to a plane mirror. By measuring the rotational speed at which returning light was blocked by adjacent teeth, he calculated $c_{\text{computed}} \approx 3.153 \times 10^8 \text{ m/s}$ (remarkably close to actual $2.998 \times 10^8 \text{ m/s}$).
Fizeau's Cogwheel Experiment
Figure 13: Fizeau's 1849 cogwheel setup for measuring the speed of light over an 8633m baseline.

3.2 Pulse Modulation (Flash ToF)

  • Operating Principle: A short, high-power laser pulse is emitted into the scene. An ultra-fast nanosecond stopwatch measures the time delay $\Delta t$ before the reflected pulse strikes the sensor.
  • Disadvantage: Sub-centimeter precision requires sub-nanosecond stopwatch electronics and high peak-power pulsed lasers, making high-resolution arrays costly.
Pulse Modulation ToF
Figure 14: Pulse modulation (Flash ToF): Measuring exact pulse time delay using nanosecond timing circuits.

3.3 Continuous Modulation (Phase ToF)

To avoid sub-nanosecond digital stopwatches, continuous modulation modulates emitted light intensity continuously using a high-frequency sinusoid (e.g., $f = 30 \text{ MHz}$).

Depth is directly proportional to the phase shift $\varphi$ measured between the emitted and returning cosine waves.

Continuous Modulation Phase ToF
Figure 15: Continuous modulation ToF: Phase shift φ between emitted and received sinusoidal light signals.

Correlation-Based Phase Measurement

The returning optical signal is demodulated by multiplying and integrating pixel charge against a reference signal $S_{ref}$ phase-locked to the emitter:

$$L_{emit} = \cos(\omega t)$$

$$L_{scene} = O + A \cos(\omega t - \varphi)$$

$$S_{ref} = \cos(\omega t - \delta)$$

Correlation-Based Phase Measurement Setup
Figure 16: Correlation-based phase measurement parameters: Ambient light O, albedo A, phase shift φ, and reference phase δ.

Measuring pixel charge under three distinct reference phase shifts ($\delta_1, \delta_2, \delta_3$) allows closed-form recovery of the unknown phase delay $\varphi$.

Phase-to-Distance Formula

Once phase shift $\varphi$ is recovered, absolute distance $d$ is given by:

$$d = c \frac{\varphi}{4\pi f}$$

Numerical Example: For modulation frequency $f = 30 \text{ MHz}$ and detected phase shift $\varphi = \pi$: $$d = (3 \times 10^8) \cdot \frac{\pi}{4\pi \cdot (30 \times 10^6)} = \frac{3 \times 10^8}{1.2 \times 10^8} = 2.5 \text{ meters}$$

3.4 Industrial Applications and Mobile Devices

  • Autonomous Vehicles (LiDAR): Mechanical rotating LiDAR systems sweep single laser beams across $360^\circ$ to generate dense 3D point clouds for autonomous navigation. Solid-state LiDAR architectures are rapidly reducing cost and size.
Google Self-Driving Car 3D Point Cloud
Figure 17: Autonomous vehicle 3D point cloud generation using scanning LiDAR / Time-of-Flight sensors.
  • Mobile Consumer Devices (Solid-State ToF): Modern smartphones and tablets integrate solid-state ToF sensor arrays. Instead of mechanical scanning, every pixel measures phase shift simultaneously, generating real-time depth maps for Augmented Reality (AR), portrait bokeh, and facial recognition.