Dual-Camera Object Perception for Low-Compute ADAS Range Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In advanced driver assistance systems (ADAS), the use of multi-scale feature extraction based on high-resolution images for object identification leads to high computing power requirements, making it difficult to deploy perception models in vehicle terminals and results in a low recall rate for distant-range objects.
Innovation Solution
Utilizing a combination of wide-angle and narrow-angle cameras to capture images, where the narrow-angle camera supplements the wide-angle camera's field of view, allowing for separate perception task models to be applied to different distance ranges, thereby improving recall rates for both close-range and distant-range objects without the need for multi-scale feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-scale feature extraction based on high-resolution images is used for object identification, then object identification capability is improved, but computing power requirements increase and recall rate for distant-range objects decreases
Solution Approach 1:
The patent divides the field of view into multiple distance ranges (first distance range for close objects, second distance range for distant objects). Different perception task models are applied to different distance ranges based on images from wide-angle and narrow-angle cameras. This segmentation allows optimized processing for each distance range, improving recall rate for distant objects while maintaining identification capability.
2Measurement precision
If multi-scale feature extraction based on high-resolution images is used, then object identification capability is improved, but computing power requirements make deployment in vehicle terminals difficult
Solution Approach 1:
The patent segments the perception task into different distance ranges and uses different camera types (wide-angle for close range, narrow-angle for distant range) with corresponding specialized perception task models. This avoids the need for computationally intensive multi-scale feature extraction on high-resolution images, reducing device complexity while maintaining identification capability.
Solution Approach 2:
The patent applies different perception task models with different characteristics to different spatial regions (distance ranges). The wide-angle camera with its model is optimized for close-range objects, while the narrow-angle camera with its model is optimized for distant-range objects. This local optimization reduces overall computing power requirements compared to a uniform high-computation approach.
3Measurement precision
If wide-angle camera is used for close-range objects, then perception precision is improved, but recall rate for distant-range objects decreases
Solution Approach 1:
The patent merges the capabilities of wide-angle and narrow-angle cameras, each with their own perception task models. The wide-angle camera handles close-range objects with high precision, while the narrow-angle camera supplements by detecting distant-range objects. The results from both cameras are combined to achieve high recall rate across the full distance range while maintaining precision for close-range objects.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Disclosed in embodiments of this disclosure are a method and apparatus for determining a perception result, a medium, and a device. The method includes: determining a first image captured by a wide-angle camera and a second image captured by a narrow-angle camera, where a field of view (FOV) of the narrow-angle camera is smaller than a FOV of the wide-angle camera; determining, based on the first image, the second image, and perception task models corresponding to distance ranges, first perception results corresponding to distance ranges; and determining an object perception result based on the first perception results corresponding to distance ranges.