Roadway Region Detection for Reliable Wrong-Way Driving Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to accurately detect wrong-way driving situations on roads, particularly at night and with varying shooting angles, and often incorrectly classify non-vehicle objects as vehicles, leading to reduced reliability and increased accidents.
Innovation Solution
A road situation detection device using a deep learning model to distinguish between roadways and sidewalks, and a single model for both daytime and nighttime conditions, capable of adapting to shooting angle changes, and utilizing tire friction sound to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep learning model is used to distinguish roadways from sidewalks, then measurement precision of vehicle detection is improved, but device complexity increases
Solution Approach 1:
The system segments the detection task into two distinct models: a region distinguishing model that first identifies roadway vs. sidewalk regions, and an abnormal situation determination model that then detects vehicles and abnormal situations within the identified roadway regions. This segmentation allows each model to specialize in a specific task, improving overall detection accuracy while managing complexity through functional decomposition.
Solution Approach 2:
The region distinguishing model acts as an intermediary between the raw image input and the vehicle detection process. It first processes the image to identify and segment roadway regions, then passes these segmented regions to the abnormal situation determination model for vehicle detection. This intermediary step improves measurement precision by ensuring vehicles are only detected in appropriate roadway contexts, while the modular architecture manages device complexity.
2Adaptability or versatility
If separate models are trained for daytime and nighttime conditions, then adaptability to different lighting conditions is improved, but device complexity and training time increase
Solution Approach 1:
The abnormal situation determination model is designed with multi-functionality to handle both daytime and nighttime detection tasks using a single unified model. The model processes images from both lighting conditions through the same architecture and training framework, eliminating the need for separate daytime and nighttime models. This universal approach maintains adaptability to different lighting conditions while reducing device complexity and model management overhead.
3Measurement precision
If training data includes all moving objects, then measurement precision is improved, but training time and computational resources increase
Solution Approach 1:
The system extracts and excludes specific moving objects (bicycles and electric scooters) from the training data of the region distinguishing model. By removing these particular object types from the training dataset, the model focuses on learning to distinguish roadways from sidewalks without being confused by two-wheeled vehicles that may appear on both roadways and sidewalks. This extraction approach improves measurement precision for vehicle detection while reducing training time and computational resources required.
4Measurement precision
If region distinguishing is performed before vehicle detection, then measurement precision of wrong-way driving detection is improved, but device complexity increases
Solution Approach 1:
The system performs preliminary action by first using the region distinguishing model to identify and segment roadway regions before the abnormal situation determination model performs vehicle detection and wrong-way driving analysis. This preliminary region identification establishes the correct spatial context and boundaries for subsequent vehicle detection, significantly improving measurement precision for detecting wrong-way driving situations. The two-stage processing pipeline manages device complexity through modular architecture where each model handles a specific subtask.
Data Source
AI summary
The present disclosure relates to a technology for determining abnormal situations (e.g., wrong-way driving) regarding the driving of a vehicle object through object recognition from a captured image, and realizes a technical configurations and embodiments for determining, with high reliability, abnormal situations (e.g., wrong-way driving) regarding the driving of a vehicle object in a roadway from a captured image.