Object Detection in Automated Driving Using Light Source Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In poorly illuminated scenes with active light sources, existing image processing systems struggle to distinguish light sources from reflections and overlapping objects, leading to incorrect object identification, particularly in automated driving applications where real training data sets are incomplete and cannot account for all potential light sources and reflections.
Innovation Solution
A method using multiple machine learning systems, including artificial neural networks, to adaptively identify light source positions, normalize images, and distinguish light sources from reflections, with synthetic generation of light source and reflection scenarios for training, allowing for efficient object identification and attribute determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If normalization is applied to adjust the image, then the surroundings become visible, but the position of light sources is no longer clearly determinable
Solution Approach 1:
The patent segments the image processing task into two distinct processing paths: one path processes the original image to detect light sources and their positions, while another path processes a normalized version of the image to identify surroundings and contexts. This segmentation allows both light source position determination and surroundings visibility to be achieved simultaneously without interference between the conflicting processing requirements.
2Extent of automation
If a machine learning system is trained to identify objects in input images, then object identification capability is achieved, but light sources are incorrectly associated with vehicles and objects leading to ghost objects
Solution Approach 1:
The patent extracts light sources from the image processing pipeline by detecting them separately and excluding them from the object identification process. A dedicated light source detection module identifies light sources in the original image, and these detected light sources are then removed or masked before being fed into the object identification machine learning system. This extraction prevents the machine learning system from incorrectly associating light sources with vehicles or other objects, thereby eliminating ghost objects while preserving automated object identification capability.
3Adaptability or versatility
If real training data is used to train the machine learning system, then the system learns from actual scenarios, but the dataset cannot include all potential light sources and reflections
Solution Approach 1:
The patent creates synthetic copies of training data by generating artificial light sources and reflections that are then integrated into real training images. These synthetic elements are superimposed onto real-world scenes to create augmented training datasets that include a comprehensive variety of light source positions, types, and reflection scenarios. This copying approach allows the machine learning system to be trained on a complete and diverse set of potential light source configurations while maintaining the realism of actual driving scenarios.
Data Source
AI summary
A computer-implemented method for reliably identifying objects in a sequence of input images received with the aid of an imaging sensor, positions of light sources in the respective input image being ascertained from the input images in each case with the aid of a first machine learning system, in particular, an artificial neural network, and objects from the sequence of input images being identified from the resulting sequence of positions of light sources, in particular, with the aid of a second machine learning system, in particular, with the aid of an artificial neural network.