Grayscale Perception Model for Autonomous Driving Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving vehicles face challenges in object identification due to color distortion in captured color images caused by tinted vehicle windows or coatings, which can negatively impact the perception process.
Innovation Solution
The use of grayscale images generated through perceptual luminance-preserving conversion from color images, with a pre-trained grayscale perception model for object identification, allowing for improved object detection and trajectory planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If color images are used for object identification, then more information is available, but color distortion occurs due to tinted windows or coatings
Solution Approach 1:
The patent extracts only the luminance information from color images by converting them to grayscale images. This removes the harmful color distortion caused by tinted windows while preserving the useful luminance information needed for object identification. The conversion process discards color channels that are distorted and retains only the brightness information that remains accurate.
Solution Approach 2:
The patent changes the parameter representation from color space (RGB) to grayscale intensity. By transforming the image data from three color channels to a single luminance channel, the system eliminates the color distortion problem while maintaining the essential visual information for detecting objects such as traffic lights, pedestrians, and other vehicles.
2Loss of information
If color images are used for object identification, then more visual information is captured, but detection accuracy decreases due to color distortion
Solution Approach 1:
The patent extracts only the reliable luminance component from the color image data, discarding the distorted color information. This extraction process maintains visual information necessary for object detection while eliminating the color distortion that reduces detection accuracy.
Solution Approach 2:
The patent transforms the image parameter representation from color space to grayscale intensity space. This parameter change improves measurement precision for object detection by using only the luminance parameter that is not affected by window tinting or anti-fog coatings.
3Object-affected harmful factors
If grayscale conversion is applied, then color distortion is reduced, but color information is lost
Solution Approach 1:
The patent converts the potentially harmful effect of color distortion into a benefit by deliberately discarding the distorted color information and using only the undistorted luminance information. The color distortion problem is transformed into an advantage where the system automatically selects the reliable grayscale channel and ignores the unreliable color channels.
Solution Approach 2:
The patent extracts only the useful luminance information from the color image, deliberately leaving out the distorted color information. This selective extraction resolves the contradiction by removing harmful color data while preserving beneficial brightness data for object identification.
Data Source
AI summary
A method, apparatus, and system for identifying objects based on grayscale images in the operation of an autonomous driving vehicle is disclosed. In one embodiment, one or more first color images are received from a camera mounted at an autonomous driving vehicle (ADV). One or more first grayscale images are generated based on the one or more first color images, which comprises converting each of the one or more first color images into one of the first grayscale images. One or more objects in the one or more first grayscale images are identified based on a pre-trained grayscale perception model. A trajectory for the ADV is planned based at least in part on the identified one or more objects. Control signals are generated to drive the ADV based on the planned trajectory.