Vehicle Occupant Behavior Recognition Using Rank-Order Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for recognizing occupant behavior in vehicles face challenges such as excessive feature extraction leading to time-consuming learning and discrimination processes, and difficulty in identifying unintended dangerous behaviors in moving images.
Innovation Solution
A behavior recognition apparatus that detects body part positions using vehicle interior images, extracts rank-order features based on the magnitude of distances between parts, and employs a discriminator learned through decision tree methods like Random Forests to accurately classify behaviors, robust against scale fluctuations and camera orientation changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If points of interest are extracted at high density and multiple types of image features are extracted for each trajectory, then the description capability of image features is improved, but the amount of extracted features becomes enormous and the learning and discrimination processes become time-consuming
Solution Approach 1:
The patent extracts only the essential feature of rank-order relationships between body part distances, rather than extracting all possible image features (HOG, HOF, MBH) at high density. This selective extraction of the most discriminative feature (relative position relationships) reduces the feature amount while maintaining behavior recognition accuracy.
Solution Approach 2:
The patent transforms the feature representation from raw coordinate data or dense trajectory features into rank-order features that represent only the relative position relationships. This parameter transformation compresses the feature space while preserving the essential behavioral information needed for discrimination.
2Reliability
If traditional feature extraction methods are used to recognize behaviors in moving images, then intended behaviors such as gestures can be effectively recognized, but unintended dangerous behaviors cannot be recognized
Solution Approach 1:
The patent creates a universal behavior recognition system that can identify both intended behaviors (gestures) and unintended dangerous behaviors (falling, unexpected movements) using the same rank-order feature extraction and discriminator approach. The system is not limited to gesture recognition but can detect any behavioral pattern through relative position changes.
Solution Approach 2:
By using rank-order features of body part distances, the system captures the essential spatial relationships that differentiate various behaviors including dangerous ones. The discriminator learns to recognize patterns in these relative position changes that indicate both intended and unintended behaviors, expanding the system's versatility.
3Device complexity
If absolute position features of body parts are used for behavior recognition, then the recognition process is simple, but the system becomes sensitive to environmental factors such as seat movement and camera orientation changes
Solution Approach 1:
The patent transforms absolute position parameters into relative position parameters (rank-order features of distances between body parts). This parameter transformation maintains the simplicity of the recognition process while eliminating sensitivity to environmental factors like seat movement and camera orientation, as these affect all body parts equally and thus cancel out in relative measurements.
Solution Approach 2:
By using relative position relationships rather than absolute positions, the system creates an equipotential reference frame that is invariant to environmental transformations. All body parts experience the same environmental shifts (camera movement, seat movement), so their relative positions remain constant, making the recognition robust.
Data Source
AI summary
Provided is a behavior recognition apparatus, including a detection unit configured to detect, based on a vehicle interior image obtained by photographing a vehicle interior, positions of a plurality of body parts of a person inside a vehicle in the vehicle interior image; a feature extraction unit configured to extract a rank-order feature which is a feature based on a rank-order of a magnitude of a distance between parts obtained by the detection unit; and a discrimination unit configured to discriminate a behavior of an occupant in the vehicle using a discriminator learned in advance and the rank-order feature extracted by the feature extraction unit. Also provided is a learning apparatus to learn the discrimination unit.