Multi-task Joint Perception Network for Traffic Road Surface Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current self-driving perception systems process tasks individually, leading to longer processing times, delays, and limited computing resources, especially when handling multiple tasks simultaneously like lane line detection and drivable area segmentation.
Innovation Solution
A multi-task joint perception network model called RSI-Net is introduced, which can simultaneously process lane line detection and drivable area segmentation tasks. This model includes a data preprocessing module, a Focus slicing module, a backbone network for feature extraction, a feature fusion module, a dilated convolution residual module, and a feature sharing module to enhance receptive fields and improve detection precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If single-task processing is used for lane line detection and drivable area segmentation, then detection precision is improved, but processing time increases and computing resources are limited
Solution Approach 1:
The patent combines lane line detection and drivable area segmentation into a single multi-task joint perception network model RSI-Net. The network shares common feature extraction modules (backbone network, feature fusion module, dilated convolution residual module) between tasks, allowing both detection tasks to be performed simultaneously rather than sequentially, thus reducing processing time while maintaining detection precision through task-specific detection heads
Solution Approach 2:
The network model is designed with multi-functionality to handle both lane line detection and drivable area segmentation tasks. The shared backbone network and feature fusion modules provide universal feature extraction capabilities that serve both tasks, while task-specific modules handle the unique requirements of each detection task, achieving efficient multi-task processing
2Productivity
If multiple tasks are processed simultaneously in a joint perception network, then processing efficiency is improved, but task-specific detection precision may be compromised
Solution Approach 1:
The network model segments the processing pipeline into shared components and task-specific components. The shared components (backbone network, feature fusion module, dilated convolution residual module) handle common feature extraction for both tasks, while separate task-specific detection heads handle lane line detection and drivable area segmentation independently, ensuring each task maintains its detection precision while benefiting from efficient joint processing
Solution Approach 2:
Different parts of the network have different functions optimized for their specific purposes. The backbone network provides general feature extraction, the feature fusion module handles multi-scale feature integration, the dilated convolution residual module enhances receptive field for context understanding, and the task-specific detection heads provide specialized processing for lane line and drivable area detection, ensuring local optimization for each task's precision requirements
Data Source
AI summary
A multi-task joint perception network model and detection method for traffic road surface information can simultaneously detect a lane line and a drivable area. A coordinate attention mechanism is integrated into a traditional feature extraction network to ensure that a feature extraction effect is enhanced while a calculated amount is not increased. In a neck network, a dilated convolution residual module is proposed to enhance performance of prediction of details by the network, and a decoder part shares features of the drivable area into lane line detection to enhance a lane line detection effect under complex road conditions. In a training stage, there is provided a alternating optimization training method to improve integral segmentation performance of the model. The multi-task joint perception network model and detection method realizes quite high accuracy and excellent speed performance in a challenging BDD100K dataset.