Hybrid CNN-RNN Road Lane Detection for Occlusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing road lane detection systems using neural networks struggle to accurately detect incomplete, partially occluded, or road lanes under drastic light changes, leading to inaccurate lane information.
Innovation Solution
A deep learning model combining convolutional neural networks (CNN) and recurrent neural networks (RNN), including bidirectional long short-term memory networks, is used to process road images, preserving image features for each pixel point and extracting spatial features of road lanes, thereby enhancing detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a neural network is applied to road lane detection, then the detection process can be automated, but the detection accuracy deteriorates for incomplete, partially-occluded, or light-changing road lanes
Solution Approach 1:
The patent combines convolutional neural networks (CNN) and recurrent neural networks (RNN) into a hybrid deep learning model. The CNN extracts spatial features from road images while the RNN captures temporal dependencies by processing sequences of images. This merging of two different neural network architectures allows the system to maintain high detection accuracy for challenging road lane scenarios while preserving automation capabilities.
Solution Approach 2:
The system performs preliminary actions by pre-processing road images to enhance feature visibility and preparing sequence data for temporal analysis. The model processes multiple consecutive images to build up contextual information before making final detection decisions, which improves accuracy for incomplete or occluded lanes by anticipating lane continuations based on temporal patterns.
2Device complexity
If conventional neural network models are used for road lane detection, then the system structure remains simple, but the ability to detect road lanes under drastically light changing environments deteriorates
Solution Approach 1:
The patent employs parameter changes by dynamically adjusting model parameters and processing parameters adapted to lighting conditions. The deep learning model learns to weight different features differently based on lighting conditions, and the sequence processing capability allows the system to adapt to temporal variations in lighting by analyzing changes across multiple frames, thereby improving versatility without excessive complexity.
Solution Approach 2:
The recurrent neural network acts as an intermediary that processes temporal sequences of images, bridging the gap between individual frame analysis and contextual understanding. This intermediary layer allows the system to compensate for poor lighting in individual frames by utilizing information from surrounding temporal context, improving adaptability while maintaining reasonable system structure.
3Adaptability or versatility
If road lane detection is performed on incomplete or partially-occluded lanes, then the system handles real-world complexity, but the detection reliability deteriorates
Solution Approach 1:
The patent applies continuity of useful action by processing sequences of images continuously to track road lanes over time. The recurrent neural network maintains temporal continuity by analyzing multiple consecutive frames, allowing the system to infer complete lane paths even when individual frames show incomplete or occluded sections. This continuous temporal processing significantly improves detection reliability for real-world complex scenarios.
Solution Approach 2:
The deep learning model incorporates feedback mechanisms where the output from processing sequence data is fed back into the system to refine detection results. The recurrent architecture allows previous temporal information to influence current detection decisions, creating a feedback loop that continuously improves detection reliability by correcting errors in incomplete or occluded lane detection through contextual validation.
Data Source
AI summary
A method and an apparatus for detecting a road lane are provided. The method includes acquiring a current road image of a road around a vehicle and inputting the current road image into a deep learning model and detecting a road lane region in the current road image based on a result outputted from the deep learning model. The deep learning model includes a first model device and a second model device. The first model device includes at least one first model subdevice which includes a convolutional neural network and a first recurrent neural network, and the second model device includes at least one second model subdevice which includes a deconvolution neural network and a second recurrent neural network.