Autonomous Path Detection Using Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous vehicle path detection systems rely on explicit lane markings, which are unreliable in environments without markings, occluded markings, or complex intersections, and require substantial post-processing, increasing computing power and reducing efficiency.
Innovation Solution
A deep learning-based system using convolutional neural networks (CNNs) for autonomous vehicles that identifies drivable paths through object detection techniques, defining paths as explicit or implied, including lane splits, cross-traffic paths, and paths without markings, reducing reliance on explicit lane markings and minimizing post-processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional computer vision algorithms with edge detection and manually engineered filters are used to identify lane markings, then the system can detect paths in environments with visible lane markings, but the system fails in environments without markings, occluded markings, or complex intersections
Solution Approach 1:
The patent transforms the detection approach by changing the fundamental parameter being detected: instead of detecting lane markings (visual features), the system directly detects drivable paths (geometric structures). This parameter transformation enables the system to work in environments where traditional marking-based detection fails, including unmarked roads, occluded markings, and complex intersections.
Solution Approach 2:
The patent inverts the conventional detection logic: rather than detecting lane markings and then inferring drivable paths through post-processing, the system directly detects drivable paths as the primary output. This inversion eliminates the need for explicit marking detection and subsequent complex post-processing steps, thereby improving both reliability and adaptability across diverse driving environments.
2Measurement precision
If deep learning is used to generate segmentation masks and perform extensive post-processing to identify lane markings, then the system can classify each pixel in an image, but the run-time increases and additional computing power is consumed
Solution Approach 1:
The patent extracts and eliminates the unnecessary post-processing steps from the conventional pipeline. Instead of generating segmentation masks and then performing extensive post-processing (filtering, smoothing, curve fitting, connected components labeling), the system directly outputs drivable path geometries in a usable format. This extraction of essential functionality removes computational overhead while maintaining detection precision.
Solution Approach 2:
The patent replaces the mechanical post-processing pipeline (multiple sequential processing steps) with a direct deep learning detection approach. The neural network is trained to output drivable path geometries directly, substituting the mechanical processing chain with an intelligent end-to-end detection system that achieves the same or better precision with significantly reduced computational cost and faster run-time.
3Ease of operation
If conventional systems perform substantial post-processing including filtering, smoothing, curve fitting, and connected components labeling to generate usable data, then the system can produce path outputs, but the computing power requirements increase and run-time is extended
Solution Approach 1:
The patent performs preliminary action by training the deep learning model to directly output drivable path geometries in a usable format during the detection phase itself. The model is pre-trained to produce results that are immediately suitable for autonomous vehicle navigation, eliminating the need for subsequent post-processing operations. This preliminary action ensures data usability is achieved without the time loss associated with extensive post-processing.
Data Source
AI summary
In various examples, a deep learning solution for path detection is implemented to generate a more abstract definition of a drivable path—without reliance on explicit lane-markings—by using a detection-based approach. Using approaches of the present disclosure, the identification of drivable paths may be possible in environments where conventional approaches are unreliable, or fail—such as where lane markings do not exist or are occluded. The deep learning solution may generate outputs that represent geometries for one or more drivable paths in an environment and confidence values corresponding to path types or classes that the geometries correspond. These outputs may be directly useable by an autonomous vehicle—such as an autonomous driving software stack—with minimal post-processing.