Vehicle Object Avoidance Communication System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition systems in vehicles often inaccurately classify detected roadway conditions, leading to unnecessary actions to avoid objects, such as braking or steering, when the object poses no harm, highlighting the need for precise classification and communication of object information to other vehicles.
Innovation Solution
A system utilizing a neural network backbone for image processing and machine learning to detect objects on a roadway, perform fine-grained classification, and communicate the object's class, risk score, and commanded actions wirelessly to other vehicles, enabling preemptive actions based on accurate data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image recognition systems classify detected roadway conditions, then objects can be identified and avoided, but classification accuracy deteriorates leading to incorrect identification of harmless objects
Solution Approach 1:
The system segments the object detection task into multiple specialized classification models. Different neural network models are trained for specific object categories (e.g., animals, debris, potholes, ice) rather than using a single general-purpose classifier. This segmentation allows each model to specialize in its domain, improving both reliability and precision for specific object types.
Solution Approach 2:
The system introduces an intermediary verification layer between detection and action. Multiple classification models vote on object identity, and a consensus mechanism determines whether an object is truly harmful. This intermediary step prevents premature actions by verifying classifications across multiple specialized models before triggering avoidance maneuvers.
2Reliability
If vehicles take avoidance actions for all detected objects, then safety is improved, but unnecessary maneuvers increase reducing driving efficiency
Solution Approach 1:
The system applies different response qualities based on the specific object characteristics. Harmless objects like paper or small debris trigger no avoidance action, while potentially dangerous objects like animals or large obstacles trigger appropriate maneuvers. This localized quality control ensures actions are taken only when necessary, maintaining both safety and driving efficiency.
Solution Approach 2:
The system uses partial action by selectively applying avoidance maneuvers only to objects that meet specific danger criteria. Rather than braking or steering for every detected object, the system evaluates object characteristics and applies avoidance actions only when the risk threshold is exceeded, preventing excessive maneuvers while maintaining adequate safety.
3Loss of information
If wireless communication is implemented to share object data between vehicles, then information availability is improved, but system complexity increases
Solution Approach 1:
The system extracts only the essential object information needed for safety (object class, location, risk level) and transmits this condensed data packet between vehicles. Rather than sharing complete sensor data or raw images, the system extracts and communicates only the critical classification results and positional information, reducing communication complexity while maintaining information availability.
Data Source
AI summary
Methods and systems for assisting a vehicle in avoiding an object in a roadway and communicating associated information to trailing vehicles. Image data is sensed by a vehicle image sensor regarding a roadway. A computing system detects an object in the roadway based on the first image data. Feature extraction is performed to extract relevant features of the object, which is fed to an object classification model to determine a class of the object. A threat analysis is performed based on the determined class of the object, outputting a risk score. The first vehicle is commanded to take an action based on the risk score. The class of the object, the risk score, and/or the commanded action taken by the vehicle is communicated to a second vehicle that is approaching the object.