Pedestrian Direction Discrimination via Weighted Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing class classifiers used in vehicles to discriminate the direction of pedestrians in images may erroneously determine the direction, leading to incorrect risk assessments of collisions, as they fail to accurately differentiate between directions, particularly when pedestrians are unaware of the vehicle or likely to enter the roadway.
Innovation Solution
A direction discrimination device that extracts feature information from images and weights it based on the likelihood of differences between directions, using a statistical learning method to minimize erroneous discrimination by prioritizing feature information from portions with significant differences between directions, such as silhouette and texture, over those with minimal differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a class classifier is used to discriminate pedestrian direction, then driving assist operation can be performed based on discriminated direction, but erroneous discrimination occurs where pedestrian direction is incorrectly determined to be largely different or opposite to actual direction
Solution Approach 1:
The patent segments the pedestrian image into multiple regions (head, body, legs, etc.) and extracts feature information from each region separately. By dividing the image into specific portions and analyzing directional features region-by-region, the system achieves more accurate overall direction discrimination while reducing erroneous determination of pedestrian orientation.
2Device complexity
If feature information from all portions of the person is weighted equally, then processing is simplified, but directional features from critical portions are not emphasized leading to higher error rates
Solution Approach 1:
The patent applies local quality by assigning different weights to feature information from different portions of the pedestrian based on their diagnostic value for direction determination. Critical regions such as the head and upper body that provide strong directional cues are given higher weights, while less informative regions receive lower weights. This selective weighting enhances measurement precision without significantly increasing processing complexity.
Data Source
AI summary
A direction discrimination device includes: an extraction unit configured to extract pieces of feature information from an image of a person, each of the pieces of feature information representing a feature of corresponding one of portions of the person; and a discrimination unit configured to discriminate a direction of the person based on the pieces of feature information. The discrimination unit is configured to weight first feature information which is feature information of a portion that is likely to have a difference in the feature between a first direction and a second direction range, more than second feature information which is feature information of a portion that is less likely to have a difference, in determination of the direction of the person. The second direction range not includes the first direction and includes a second direction opposite to the first direction.