Child Collision Risk Detection for Autonomous Vehicle Vision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately anticipating and responding to the unpredictable behavior of children, as they lack human intuition and experience, making it difficult to effectively determine collision risks.
Innovation Solution
A method that uses image data to determine the height, age, and attributes of pedestrians, classifying them as critical or non-critical, and assess their position relative to the vehicle, thereby accurately identifying collision risks and triggering appropriate mitigation maneuvers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If autonomous vehicles use standard pedestrian detection methods, then detection speed is maintained, but detection accuracy for children is insufficient
Solution Approach 1:
The pedestrian detection process is segmented into multiple stages: initial detection, attribute classification (divided into body attributes, object attributes, and state attributes), and risk assessment. This segmentation allows the system to focus computational resources on critical attributes while maintaining overall detection speed.
Solution Approach 2:
The system applies different levels of analysis to different attributes of detected pedestrians. Critical attributes (such as height, presence of stroller, backpack status) undergo detailed classification, while non-critical attributes receive minimal processing. This local quality approach optimizes the balance between detection accuracy and system complexity.
2Measurement precision
If the system analyzes all pedestrian attributes in detail, then collision risk detection accuracy improves, but processing time increases
Solution Approach 1:
The system performs partial analysis by selectively evaluating only the attributes necessary for collision risk assessment. Not all pedestrian attributes are analyzed in full detail - only those classified as critical based on their relevance to collision risk. This partial action approach maintains detection accuracy while reducing processing time.
Solution Approach 2:
The system performs preliminary classification of attributes into critical and non-critical categories before conducting detailed analysis. This preliminary action allows the system to prepare for efficient processing by identifying which attributes require detailed evaluation, thereby reducing overall processing time while maintaining accuracy.
3Reliability
If the system triggers collision mitigation maneuvers for all detected children, then safety is maximized, but false alarms increase
Solution Approach 1:
The system uses feedback from multiple attribute classifications (body attributes, object attributes, state attributes) to dynamically adjust collision risk assessment. The feedback loop evaluates whether detected children actually pose a collision risk based on their attributes and state, allowing the system to distinguish between true risks and false alarms while maintaining high safety standards.
Solution Approach 2:
The system changes the assessment parameters dynamically based on detected attributes. Instead of applying a fixed threshold for all children, the system adjusts risk parameters according to specific attribute combinations (e.g., a child with a stroller near the road vs. a child playing in a park). This parameter adaptation reduces false alarms while maintaining reliability.
4Reliability
If the system uses multiple classification criteria for pedestrians, then detection reliability improves, but system complexity increases
Solution Approach 1:
The classification system is segmented into three distinct attribute categories: body attributes (height, age estimation), object attributes (stroller, backpack, ball), and state attributes (moving, standing, playing). This segmentation allows each category to be processed independently with specialized algorithms, improving reliability while managing complexity through modular design.
Solution Approach 2:
The classification framework is designed to be universal, handling multiple types of pedestrians and scenarios through a single integrated system. The same attribute classification structure applies to all detected pedestrians regardless of their specific characteristics, reducing the need for multiple specialized systems and thereby managing complexity while maintaining high detection reliability.
Data Source
Figure 1~2
Figure 3~4
AI summary
The disclosure relates to a method for determining a collision risk between an autonomous vehicle (12) and a child (16a, 16b). The method comprises receiving image data (I) comprising a representation of at least one pedestrian (P). The method further comprises determining a height (H) of the at least one pedestrian (P) and/or determining an age of the at least one pedestrian (P). Moreover, the method comprises determining a pedestrian attribute (PA) characterizing the at least one pedestrian (P) and a pedestrian position indicator (PI) characterizing a position of the pedestrian (P). Based thereon, a presence of a collision risk is determined if the determined height (H) is inferior to a predefined height threshold and/or if the determined age is inferior to a predefined age threshold, if the pedestrian attribute (PA) is classified as critical, and if the pedestrian position indicator (PI) is classified as critical. Otherwise an absence of a collision risk is determined. Moreover, a method for operating an autonomous vehicle (12) is presented. Furthermore, a data processing apparatus (18), a computer program (26), and a computer-readable storage medium (24) are shown.