Object Detection Using Body Part Segmentation and 3D Position Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition techniques fail to accurately detect objects associated with a person when parts of the body are not fully visible in the captured image, such as when an obstacle obstructs the arm, leading to incomplete recognition of objects like bags.
Innovation Solution
An information processing apparatus that detects multiple body parts and estimates the position of target objects using positional relationships between detected parts, allowing for robust prediction of object locations even if some parts are not visible.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object recognition is performed by associating objects with specific body parts (one-to-one basis), then recognition accuracy is improved when parts are visible, but recognition fails when parts are occluded or outside imaging range
Solution Approach 1:
The patent segments the body into multiple parts (head, torso, arms, legs) and detects them independently. By dividing the body recognition task into multiple part detection tasks, the system can identify objects even when only some body parts are visible, resolving the contradiction between accuracy and reliability under occlusion.
Solution Approach 2:
The patent transitions from two-dimensional image space to three-dimensional space representation by constructing a 3D body model from detected body parts. This dimensional transformation allows the system to infer the position of occluded parts and associated objects by considering spatial relationships in 3D space, thereby maintaining recognition reliability even when parts are occluded.
2Productivity
If the imaging range is limited to capture specific body parts, then detection speed is improved, but completeness of object detection deteriorates when the whole body is not included
Solution Approach 1:
The system segments the detection task by focusing on key body parts that are most likely to be visible and associated with target objects. This selective segmentation allows faster detection of relevant objects without requiring complete body visibility, thus improving detection speed while maintaining detection completeness for objects of interest.
Solution Approach 2:
The patent performs preliminary detection of body parts that are most likely to be visible and associated with target objects. By pre-identifying these key parts and their typical object associations, the system can quickly detect objects without waiting for complete body detection, thereby improving detection speed while maintaining completeness.
3Reliability
If multiple body parts are detected and processed, then object detection robustness is improved, but computational complexity increases
Solution Approach 1:
The patent segments the computational task into independent body part detection modules. Each module processes one body part independently, allowing parallel computation and reducing overall computational complexity while maintaining robustness through multiple detection points.
Solution Approach 2:
The system performs detection on a partial set of body parts that are most relevant to object detection, rather than processing all possible body parts equally. This partial action approach reduces computational complexity while maintaining sufficient robustness for practical object detection applications.
Data Source
AI summary
An information processing apparatus (2000) detects an estimation position (24), which is a position where a target object is estimated to be present, from a captured image (20) generated by a camera (10). The target object is a belonging of a person. The information processing apparatus (2000) detects a plurality of parts (22) constituting a body of a person included in the captured image (20). For example, the parts (22) are joints. The information processing apparatus (2000) detects the estimation position (24) of the target object in the captured image (20) using information relating to a position of each of a plurality of the detected parts (22).