Evasive Maneuver Data Collection for Object Detection Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection models for autonomous vehicles face challenges in detecting rare or unexpected objects like road debris, due to the lack of representative training data.
Innovation Solution
A method is introduced where vehicles collect and store sensor data when evasive maneuvers are detected, even if the object detection model fails to identify an object, to enhance the training data for object detection models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional object detection models are trained only on commonly encountered objects, then the model performs well on common objects, but it fails to detect rare or unexpected objects like road debris
Solution Approach 1:
The system proactively collects sensor data during evasive maneuvers before training data is needed, preparing a dataset of rare objects in advance. This preliminary data collection enables the model to eventually detect rare objects without requiring extensive manual data gathering campaigns.
Solution Approach 2:
The automated driving system itself generates training data by detecting its own evasive maneuvers and collecting corresponding sensor data. This self-service approach allows the system to automatically improve its own capabilities without external intervention, continuously expanding its ability to detect rare objects.
2Quantity of substance
If manual data collection methods are used to gather training data for rare objects, then some training data can be obtained, but the process is extremely time-consuming and inefficient
Solution Approach 1:
The patent replaces manual mechanical data collection methods with an automated electronic system that detects evasive maneuvers and automatically collects sensor data. This substitution eliminates the need for human operators to manually review and collect data, reducing collection time from months to minutes while dramatically increasing the quantity of training data obtained.
3Quantity of substance
If the object detection model stores all sensor data for potential training, then comprehensive training data is available, but storage resources are wasted on common objects that already have sufficient training data
Solution Approach 1:
The system extracts only the relevant subset of sensor data by identifying evasive maneuver events and collecting data specifically associated with those events. This extraction approach filters out the vast majority of routine driving data that would be redundant for training, storing only the critical portions that contain rare or unexpected objects needed to improve model performance.
Data Source
AI summary
The present invention relates to a method for collecting data for subsequent training of an object detection model of an automated driving system. The method includes, in response to detecting an evasive maneuver of the vehicle or a further road user, obtaining sensor data pertaining to a scene at which the evasive maneuver was detected; determining by the object detection model, whether an object is detected in the scene based on the sensor data; and in response to no object being detected, storing the sensor data for subsequent training of the object detection model. The present invention further relates to a method performed in a server.