Dynamic Vision Sensor Surface Reconstruction with Structured Light
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor technologies for motion planning in mobile robots face challenges in efficiently detecting and reconstructing terrain surfaces at low computational cost, as they often consume high power and have limited spatial resolution due to the use of 3D scanners or passive vision systems.
Innovation Solution
A method utilizing spatially structured light with temporally varying intensity, detected by a dynamic vision sensor (DVS) that generates asynchronous address-events based on photocurrent changes, allowing for efficient 3D surface reconstruction with reduced computational overhead and flexible pulsing frequencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D scanners such as Microsoft Kinect or LIDAR devices are used for terrain detection and reconstruction, then measurement precision and surface reconstruction capability are improved, but power consumption increases to several watts and sample rate is limited to tens of Hertz
Solution Approach 1:
The patent replaces traditional active 3D scanning systems (Kinect, LIDAR) with a passive vision system using a dynamic vision sensor (DVS). The DVS captures terrain information through natural light reflection without requiring active illumination, dramatically reducing power consumption while maintaining measurement capability. The event-based asynchronous output of DVS replaces frame-based capture, enabling higher effective sample rates for motion detection.
Solution Approach 2:
The system uses the robot's own motion through the terrain to provide the necessary relative movement for surface reconstruction. Instead of requiring high-speed active scanning, the passive vision system leverages the natural motion of the robot to gather sufficient data points for 3D reconstruction, reducing the need for high-power active sensors.
2Measurement precision
If traditional 3D scanners are used for terrain detection, then surface reconstruction capability is improved, but sample rate is limited to tens of Hertz reducing productivity
Solution Approach 1:
The patent employs a dynamic vision sensor (DVS) that operates asynchronously without fixed frame rates. The sensor continuously generates events based on changes in the visual scene, effectively providing unlimited sample rates for dynamic terrain detection. This dynamic approach replaces the static frame-based capture of traditional sensors, enabling real-time reconstruction of fast-moving terrain.
Solution Approach 2:
The system uses periodic modulation of the light source (when present) to trigger synchronized capture events, creating effective sampling at the modulation frequency rather than being limited by sensor frame rates. This allows the system to achieve higher effective sample rates by modulating the illumination at frequencies beyond what traditional continuous capture could achieve.
3Use of energy by moving object
If passive vision systems are used for terrain detection, then power consumption is reduced, but spatial resolution is limited due to restriction to a small set of feature points
Solution Approach 1:
The patent applies structured light patterns (such as laser lines or grids) to illuminate specific regions of the terrain, creating high-contrast features that the passive vision sensor can detect with high precision. This localized illumination enhances the information content in specific areas without requiring high power across the entire field of view, maintaining spatial resolution while keeping power consumption low.
Solution Approach 2:
The system changes the spatial and temporal parameters of illumination by using modulated light patterns rather than continuous illumination. By encoding information in the temporal modulation and spatial structure of light, the system extracts more information from each captured event, effectively increasing spatial resolution without increasing power consumption proportionally.
4Measurement precision
If high frame rate capture is used to handle fast motions, then measurement precision for dynamic scenes is improved, but computational overhead and power consumption increase
Solution Approach 1:
The patent extracts only the essential information from the visual scene by using event-based capture that records only changes rather than complete frames. This extraction approach removes redundant data (unchanged regions) while preserving critical motion information, significantly reducing computational overhead for processing dynamic scenes while maintaining accuracy for moving terrain.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables faster and more detailed surface reconstruction with lower power consumption, capable of handling fast motions and varying brightness conditions, while maintaining high spatial resolution and reducing redundant data processing.
Implementation Method 1
each pixel generates a signal (e.g. monotonically) related to a photocurrent generated by the pixel, which photocurrent is proportional to the intensity of the light impinging on the respective pixel
Data Source
Figure 1A~2
Figure 3A~4
Figure 5A~5B
AI summary
The present invention relates to a method for detecting and particularly reconstructing a surface (40), wherein the surface (40) is illuminated with spatially structured light, such that said light illuminates an area (31 ) of said surface (40) from which said light is reflected back, wherein said light comprises a temporarily varying intensity (I) in the form of successive light modulation patterns (32), wherein back- reflected light is detected by means of an optical sensor (20) that comprises a plurality of pixels, wherein a pixel coordinate (u,v) is associated to each pixel, and wherein each pixel generates a photocurrent being proportional to the intensity of the light impinging on the respective pixel, and computes a signal related to said photocurrent, and wherein each pixel outputs an address-event (Ev) merely when the respective signal due to the light impinging on the respective pixel increases by an amount being larger than a first threshold (ΘON) or decreases by an amount being larger than a second threshold (ΘOFF) since the last address-event (Ev) from the respective pixel, and wherein as a current image of said area, also denoted as laser stripe in case spatially structured light in the form of a plane/sheet of laser light is used, pixel coordinates (u,v) of address-events (Ev') caused by light of the last light modulation pattern (32) back-reflected from said area (31) onto the optical sensor (20) are determined. Furthermore, the invention relates to a corresponding computer program and a corresponding system.