Pedestrian Article Detection Using Deep Learning Without Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pedestrian detection methods in surveillance environments face challenges when pedestrians are partially covered, making it difficult to accurately detect clothing areas and other items, and are prone to noise-related inaccuracies due to complex segmentation processes.
Innovation Solution
A pedestrian article detection apparatus and method that uses a deep learning model directly in the detected pedestrian area to identify and categorize articles, eliminating the need for clothing area determination and contour-based segmentation, thereby enhancing detection accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If area segmentation is performed according to contour of human body and classification model is used for detection in segmented areas, then detection can be performed for specific body parts, but the process is relatively complicated and prone to being affected by noises which may result in inaccurate detection results
Solution Approach 1:
The patent extracts and removes the complex contour-based segmentation process from the detection system. Instead of segmenting the pedestrian area into upper and lower body regions using contour detection, the invention directly applies the deep learning model to the entire pedestrian area, eliminating the segmentation step that causes complexity and noise sensitivity while maintaining detection accuracy for different body parts
Solution Approach 2:
The patent applies segmentation at the model level rather than at the image processing level. The deep learning model is trained to simultaneously detect multiple types of articles (upper clothes, lower clothes, bags, decorations) in one unified detection process, avoiding the need for manual image segmentation while achieving comprehensive article detection
2Measurement precision
If classification model is used in determined upper clothes area and lower clothes area to identify upper clothes and lower clothes, then specific clothing items can be identified, but the method cannot detect other items than clothes and fails when pedestrian is partially covered
Solution Approach 1:
The patent implements a universal detection model that can identify multiple types of articles simultaneously - not only upper and lower clothes but also bags, decorations, and other items carried by pedestrians. The deep learning model is designed with multi-functionality to handle diverse article types within the pedestrian area without requiring separate detection processes for each item category
Solution Approach 2:
The patent performs preliminary detection of the entire pedestrian area first, then uses the deep learning model to identify all types of articles within that area in one step. This preliminary action of detecting the full pedestrian region ensures that even partially covered items are included in the detection scope, and the model can identify various article types regardless of their position or visibility
Data Source
AI summary
Embodiments of this disclosure provide an apparatus and method. As target detection is performed in the detected pedestrian area by directly using the deep learning model, various article belonging to a pedestrian can be accurately detected in an input image.