Vision Sensor In-Pixel Current Integration for Depth Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vision systems for depth estimation, such as lidars and time-of-flight cameras, require sub-nanosecond resolution and complex setups, while passive systems like stereo vision need substantial computational resources and sufficient lighting, and active systems with structured lighting face limitations in spatial resolution and are prone to motion artifacts.
Innovation Solution
A vision sensor with an array of pixels, each comprising a photosensor, a current source that maintains a constant current level, an integrator, and a counter that measures time when the integration difference reaches a threshold, optimized for time-multiplexed structured lighting to minimize motion artifacts and reduce computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If lidars or time-of-flight cameras are used for depth measurement, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent replaces complex mechanical time-of-flight measurement systems with a vision-based system that uses structured lighting patterns and image processing. Instead of using expensive lidars or complex time-of-flight cameras requiring sub-nanosecond resolution, the invention uses a standard camera capturing images at different lighting intensities to compute depth through ratio-based calculations, thereby substituting mechanical precision requirements with computational methods.
Solution Approach 2:
The patent uses a projected light pattern that creates a virtual copy of the scene's geometry on the image sensor. By projecting structured lighting and capturing the reflected pattern, the system creates an optical copy of depth information that can be processed computationally, avoiding the need for direct physical measurement devices like lidars.
2Device complexity
If stereo vision or structure-from-motion is used, then device complexity is reduced, but computational resources and lighting requirements increase
Solution Approach 1:
The patent employs periodic modulation of light intensity across different regions of the scene. By systematically varying the intensity of projected light patterns in a controlled sequence and capturing multiple images with different lighting conditions, the system enables depth computation through intensity ratios, reducing reliance on complex stereo matching or structure-from-motion algorithms while maintaining automated depth mapping.
3Measurement precision
If structured lighting with unique features is used, then depth estimation is enabled, but spatial resolution is limited
Solution Approach 1:
The patent applies different lighting intensities to different local regions of the scene simultaneously. By encoding spatial information through local intensity variations rather than requiring unique global features, the system achieves both depth estimation and high spatial resolution. Each pixel's depth is determined by the intensity ratio of captured light, allowing dense depth maps without sacrificing spatial detail.
4Area of moving object
If time-multiplexed patterns are used, then spatial resolution is improved, but motion artifacts increase
Solution Approach 1:
The patent uses simultaneous periodic modulation of light intensity across multiple spatial regions rather than sequential time-multiplexed patterns. By capturing all spatial information in a single image through intensity-encoded patterns, the system eliminates temporal delays between pattern projections, thereby preventing motion artifacts while maintaining high spatial resolution for depth estimation.
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 vision sensor achieves high signal-to-noise ratio and precise temporal resolution through in-pixel background subtraction and illumination change detection, enabling efficient depth estimation with reduced computational complexity and motion artifacts.
Implementation Method 1
a photosensor which is configured to output a current proportional to the intensity of light which is incident on the photosensor
Data Source
AI summary
According to the present invention there is provided a vision sensor comprising, an array of pixels comprising rows and columns of pixels, wherein each pixel in the array comprises, a photosensor which is configured to output a current proportional to the intensity of light which is incident on the photosensor; a current source which is configured such that it can output a current which has a constant current level which is equal to the current level of the current output by the photosensor at a selected first instant in time, and can maintain that constant current level even if the level of the current output from the photosensor changes after said selected first instant in time; an integrator which is configured to integrate the difference between the level of current output by the current source and the level of current output by the photosensor, after the selected first instant in time; wherein the vision sensor further comprises a counter which can measure time, wherein the counter is configured such that it can begin to measure time at the selected first instant; and wherein each pixel in the array further comprises a storage means which can store the value on the counter at a second instant in time, the second instant in time being the instant when the integration of the difference between the level of current output by the current source and the level of current output by the photosensor of that pixel reaches a predefined threshold level. There is further provided a corresponding method of vision sensing, and a depth sensor assembly which comprises the vision sensor.