Occlusion-Based Obstacle Detection for Mono Camera Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional learning-based object recognition algorithms for autonomous driving systems struggle to reliably detect hazardous objects without sufficient human-annotated examples, and require stereo cameras, limiting their applicability to mono camera systems.
Innovation Solution
A method using self-supervised training to determine occlusion labels from image data, enabling the detection of obstacles in both stereo and mono camera systems by estimating optical flow and training a machine learning model to generate occlusion maps, which can be used for obstacle detection without requiring movement information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional learning-based object recognition algorithms are used, then obstacle detection can be performed, but sufficient human-annotated examples are required which are practically impossible to obtain for all hazardous object variants
Solution Approach 1:
The system performs self-supervised learning by automatically generating occlusion labels from optical flow estimates without requiring manual human annotation. The algorithm serves itself by creating its own training labels from the visual data, eliminating the need for extensive manually labeled datasets while maintaining detection reliability
Solution Approach 2:
The system performs preliminary occlusion detection and label generation before final obstacle classification. By first identifying occluded regions through optical flow analysis and then using these pre-generated labels for training, the system prepares training data automatically without requiring manual annotation of all hazardous objects
2Adaptability or versatility
If deterministic algorithms with heuristics are used to detect dangerous objects, then some unexpected obstacles can be detected, but the approach does not generalize to different scenarios and requires a stereo camera
Solution Approach 1:
The occlusion detection algorithm is designed to be universal and works with both stereo and mono camera systems. By formulating the occlusion problem in terms of optical flow and photometric error that can be computed from single images, the method achieves multi-functionality across different camera configurations without requiring stereo vision
Solution Approach 2:
The system replaces the mechanical/stereo vision requirement with a computational approach using optical flow estimation from mono camera sequences. Instead of relying on stereo geometry, the method substitutes computational optical flow analysis to achieve the same occlusion detection goal, reducing hardware complexity
3Measurement precision
If motion information is used for occlusion detection, then occlusion labels can be determined, but the approach requires image sequences rather than individual images
Solution Approach 1:
The system performs preliminary training using image sequences to learn optical flow patterns, then applies this learned knowledge to detect occlusions in individual images. The motion information is used in advance during training to establish the model, which can then operate on single images without requiring motion input during deployment
Data Source
AI summary
The invention relates to method (100) for detecting at least one obstacle in an automated and/or at least semi-autonomous driving system (60), said method comprising the following steps:providing (101) image data, wherein the image data are specific to a recording of an environment of the driving system (60),performing (102) an evaluation of the image data provided, wherein the evaluation takes place based on an application of a machine learning model (50), by means of which an occlusion label is determined for at least one occlusion of the environment,performing (103) the detection of the at least one obstacle on the basis of the occlusion label determined.