Vehicle Collision Warning Device Using Area of Interest Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current advanced driver assistance systems (ADAS) face challenges in efficiently processing and predicting the trajectories of multiple detected objects, leading to increased processing power requirements and unnecessary tracking of irrelevant objects.
Innovation Solution
A driving action recommendation and collision warning device that focuses on extracting an area of interest from images using structured and unstructured data, allowing for more efficient processing and prediction by filtering out irrelevant objects and adapting to changing traffic scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all detected objects are tracked and processed, then collision detection coverage is improved, but processing power requirements increase
Solution Approach 1:
The system extracts only the relevant area of interest from the complete sensor field of view, separating useful information (objects within AOI) from irrelevant information (objects outside AOI). This extraction principle reduces the number of objects that need to be tracked and processed, thereby lowering processing power requirements while maintaining collision detection coverage for relevant objects.
Solution Approach 2:
The system segments the sensor field of view into distinct regions: the area of interest (AOI) and areas outside AOI. This segmentation allows differential processing where only objects within the AOI undergo full tracking and collision analysis, while objects outside the AOI are either ignored or processed with reduced complexity, resolving the contradiction between comprehensive detection and processing efficiency.
2Reliability
If all detected objects are tracked, then object detection completeness is improved, but processing efficiency deteriorates
Solution Approach 1:
The system extracts the area of interest from the complete sensor data, isolating only those regions that require full processing attention. This extraction enables the system to maintain complete detection of relevant objects while excluding irrelevant objects from the tracking pipeline, thereby improving processing efficiency without sacrificing detection completeness for important targets.
Solution Approach 2:
By segmenting the processing workload based on spatial location (inside vs. outside AOI), the system applies different processing intensities to different object sets. Objects within the AOI receive complete tracking and analysis, ensuring detection completeness, while objects outside the AOI are processed minimally or not at all, thereby improving overall processing efficiency.
3Measurement precision
If trajectory prediction is performed for all objects, then prediction accuracy is improved, but processing load increases
Solution Approach 1:
The system extracts the area of interest from the complete sensor field, identifying only those spatial regions where trajectory prediction is necessary for collision avoidance. By performing computationally intensive trajectory predictions only for objects within the AOI and excluding objects outside the AOI, the system maintains prediction accuracy for relevant targets while significantly reducing the overall processing load.
Solution Approach 2:
The system segments the trajectory prediction task by spatial location, applying full prediction algorithms only to objects within the AOI while excluding or simplifying processing for objects outside the AOI. This segmentation preserves prediction accuracy for collision-relevant objects while reducing the cumulative processing load across all detected objects.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
There is described a driving action recommendation and collision warning device for a vehicle comprising a first onboard sensor 10 configured to receive images of an area in the vicinity of the vehicle from an image sensor and to detect obstacles in the received images; a second onboard sensor 20 configured to acquire structured data of the vehicle; a computing unit 30 configured to generate an image by converting the images received from an image sensor into a schematic representation comprising the obstacles detected by the first onboard sensor 10, an area of interest decision unit 40 configured to extract an area of interest from the image generated by the computing unit 30; a driver action prediction unit 50 unit configured to predict a possible future event based on the extracted area of interest and a driver action prediction model, an action recommendation unit 60 configured to output a driving action recommendation and/or a warning to a driver of the vehicle or a person in the vicinity of the vehicle when a predefined event is predicted.