Monocular Vision Depth Sensing via Super-Pixel Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle auto-pilot and navigation systems rely on multi-view identification, which requires multiple image capturing devices to analyze scenes for depth detection and object identification, leading to inefficiencies in processing speed and device burden.
Innovation Solution
A scene analyzing method using super pixel clustering recognition and spiral sampling monocular vision spatial depth sensing technology, allowing for object recognition with a single monocular vision device by capturing and analyzing scene information over time to classify targets as type A or type B based on their movement relative to the device, optimizing processing speed and reducing device burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple image capturing devices are used for multi-view identification, then depth detection and object identification accuracy are improved, but device complexity and processing burden increase
Solution Approach 1:
The patent segments the image processing task by dividing the image into super-pixels (groups of pixels with similar characteristics) and processing these segments rather than individual pixels. This reduces the data volume while preserving depth information, allowing accurate depth detection with a single camera.
Solution Approach 2:
The patent introduces temporal dimension by capturing images at multiple time points and using spiral sampling patterns to create virtual multi-view information from a single camera. This transforms a 2D monocular problem into a 3D+T (space+time) problem, enabling depth detection without multiple simultaneous cameras.
2Measurement precision
If multiple image capturing devices are used for multi-view identification, then object identification accuracy is improved, but processing speed decreases due to increased data volume
Solution Approach 1:
The patent extracts only the essential features needed for object identification by using spiral sampling to select specific super-pixels rather than processing all pixels. This extraction approach maintains identification accuracy while significantly reducing processing load and increasing speed.
Solution Approach 2:
The patent changes the processing parameters by working with super-pixel level data instead of pixel-level data, and by using temporal sequences of monocular images to compensate for the reduced spatial information. This parameter change enables faster processing while maintaining accuracy.
3Device complexity
If a monocular vision device is used, then device complexity is reduced, but depth sensing capability is weakened
Solution Approach 1:
The patent uses periodic action by capturing images at multiple time points in a sequence and applying spiral sampling patterns. This temporal periodicity compensates for the lack of spatial baselines, enabling the single camera to infer depth information through time-varying observations.
Solution Approach 2:
The patent introduces super-pixel clustering and spiral sampling as intermediary processing steps that transform monocular 2D image data into depth-aware representations. These intermediaries bridge the gap between simple monocular input and accurate depth output without requiring additional sensors.
Data Source
AI summary
A scene analysis method is applied to a visual navigation device. The scene analysis method includes the steps of capturing scene information at different times based upon the field of view of an image capturing device; analyzing different targets existing in the captured scene information; comparing each one of the targets captured at different times to classify the target, wherein the target that is in a specific region in the field of view and moving toward the image capturing device is classified as a type A target; otherwise the target is classified as a type B target. When the visual navigation is installed on a vehicle, a protection procedure such as a procedure of avoiding an obstacle is initiated when a type A target is detected, so that the vehicle is automatically prevented from potential accidents.