Photon Timing Sketching for Low-Power ToF Depth Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-rate, high-resolution, low-power time of flight (ToF) image sensors face data processing bottlenecks due to large data volumes, leading to computational burdens and accuracy compromises in existing methods for depth reconstruction.
Innovation Solution
A sensor device that generates a compressed representation of photon detection events using feature functions to preserve signal information while suppressing background noise, allowing for reduced storage and computational resources without sacrificing depth precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high spatial resolution and fine time resolution are used in ToF imaging, then depth precision is improved, but data volume increases causing computational burden and power consumption to increase
Solution Approach 1:
The patent extracts only the essential features from the full photon detection data by computing histograms that capture the dominant depth information. Instead of processing all raw photon timing data, the system extracts histogram representations that preserve depth precision while discarding redundant information, thereby reducing computational burden.
Solution Approach 2:
The patent transforms the data representation from raw photon timing events to histogram-based probability distributions. This parameter transformation allows the system to maintain depth measurement precision while working with compressed data structures that require less computational resources for processing.
2Measurement precision
If high spatial resolution and fine time resolution are used in ToF imaging, then depth precision is improved, but data volume increases causing storage requirements and power consumption to increase
Solution Approach 1:
The patent extracts only the essential features from the full photon detection data by computing histograms that capture the dominant depth information. Instead of storing all raw photon timing data, the system extracts histogram representations that preserve depth precision while discarding redundant information, thereby reducing storage requirements.
Solution Approach 2:
Instead of compressing data after acquisition, the patent inverts the approach by directly acquiring compressed histogram data through the sensor architecture. The sensor performs histogramming in-pixel, converting raw photon events into compressed representations at the point of detection, thereby reducing data volume from the outset.
3Productivity
If known compression techniques are used to reduce data rates, then data transfer burden is reduced, but computation accuracy or resolution is compromised
Solution Approach 1:
The patent performs histogram computation and data compression as a preliminary action within the sensor device before data leaves the chip. By pre-processing the photon detection data into histogram representations in-pixel, the system reduces data transfer burden while preserving the statistical information needed for accurate depth reconstruction, avoiding later compromises in accuracy.
4Reliability
If more photons per pixel are detected, then signal-to-noise ratio is improved, but data volume increases causing processing bottleneck
Solution Approach 1:
The patent extracts histogram representations from photon detection data that capture the essential signal information while discarding redundant details. This extraction process maintains signal-to-noise ratio by preserving the statistical distribution of photon arrivals, while simultaneously reducing data volume to eliminate processing bottlenecks.
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
The solution enables efficient data transfer and processing, reducing computational resources by up to 30 times while maintaining high accuracy in 3D imaging, allowing for fast frame rates and low power consumption.
Implementation Method 1
Each pixel device may comprise a single photon avalanche diode (SPAD)
Data Source
AI summary
A sensor device (10) for photon-based imaging comprising one or more photon detectors configured to produce a plurality of photon detection signals in response to a plurality of photon detection events, wherein each photon detection event has a corresponding detection time and wherein the detection times of the plurality of photon detection events are distributed in accordance with a distribution over time; and processing circuitry configured to perform a sketching process using timing information of the plurality of photon detection events to obtain a compressed representation of the distribution over time, wherein the sketching process comprises: generating a plurality of feature values based on the timing information of the plurality of photon detection events using one or more feature functions; combining the generated plurality of feature values to obtain the compressed representation of the distribution over time, wherein the feature functions have one or more properties such that combining the plurality of feature values generated using the one or more feature functions preserves signal information and/or supresses background information and/or distinguishes signal information from background information in the compressed representation, wherein the compressed representation is such that at least one or more desired parameters of the distribution over time can be estimated by performing a parameter estimation process using the compressed representation, wherein the parameter estimation process is based on a model of the distribution over time.