Oriented Bounding Box Detection for Road Safety Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current road safety assessment methods are manual, complex, and costly, and existing object detection algorithms, such as YOLO, struggle with accurately detecting oriented objects like curbs and lane markings due to their inability to account for orientation, leading to reduced accuracy in road safety evaluations.
Innovation Solution
A method using image data from environmental sensors to determine bounding boxes with two orientation descriptors (sin(2θ) and cos(2θ) to accurately classify oriented objects, enabling the computation of safety indicators for improved road safety assessments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual road surveying and data collection methods are used, then comprehensive road safety data can be collected, but the process is complex, time-consuming, and costly
Solution Approach 1:
The patent replaces manual mechanical surveying methods with an automated image processing system using computers and algorithms. The system automatically captures images of road segments and processes them to extract safety-relevant information, eliminating the need for manual measurement and surveying operations while maintaining data accuracy.
Solution Approach 2:
The system enables self-service data collection by automatically processing images to identify road features, hazards, and safety indicators without requiring manual intervention. The automated algorithm analyzes images independently to extract comprehensive safety data, allowing continuous operation without human operators.
2Productivity
If conventional object detection algorithms like YOLO are used, then fast processing is achieved, but accuracy for detecting oriented objects like curbs and lane markings is reduced
Solution Approach 1:
The patent changes the parameter representation from standard bounding boxes to oriented bounding boxes that include orientation angles. This parameter transformation enables the system to accurately represent and detect oriented objects like curbs and lane markings while maintaining processing efficiency through optimized algorithmic approaches.
Solution Approach 2:
The system adds an orientation dimension to the traditional two-dimensional bounding box representation. By incorporating orientation angles into the bounding box parameters, the system transitions from simple rectangular detection to oriented object detection, enabling accurate identification of curved and angled road features without sacrificing processing speed.
3Device complexity
If single orientation angle parameters are used in object detection, then computational complexity is reduced, but discontinuity problems arise affecting detection accuracy
Solution Approach 1:
The patent introduces an intermediary representation using oriented bounding boxes with orientation angles as a mediator between the image data and object detection algorithms. This intermediary structure enables continuous and accurate representation of oriented objects, resolving discontinuity issues while maintaining manageable computational complexity through standardized orientation parameter handling.
Data Source
Figure 1
Figure 2
Figure 3a~3b
AI summary
According to a method for road safety assessment, image data (7) representing a road segment (5) is received by a computing unit (3). A set of bounding boxes (14) is determined by the computing unit (3) based on the image data (7), each bounding box (14) representing an object (6a, 6b), wherein for each of the bounding boxes (14) a set of two orientation descriptors (21b) specifying a two-dimensional orientation of the respective bounding box (14) is determined. A respective object class is determined for each of the objects (6a, 6b) represented by the set of bounding boxes and at least one safety indicator (15) is computed for the road segment (5) by the computing unit (3) depending on the determined object classes.