Multi-Headed Person and Face Detection Model for Whole Person Association

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing human detection systems in imagery face issues with false positives and false negatives, requiring separate face and person detectors that consume significant resources and often produce conflicting information, lacking a direct association method to determine if detections refer to one or multiple humans.

Innovation Solution

A computer system that uses a multi-headed person and face detection model to associate face and person detections, adjusting confidence scores and thresholds based on alignment and pose data to improve detection accuracy, and hallucinates face detections when necessary to reduce false negatives.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If separate face detector and person detector are used, then detection coverage is improved, but resource consumption increases significantly

Engineering Contradiction:
Improvedetection coverageVSAvoidresource consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent combines face detection and person detection into a unified detection framework where a single detector performs both functions. The detector outputs both face detection results and person detection results from the same processing pipeline, eliminating the need for two separate detectors and reducing computational resource consumption while maintaining comprehensive detection coverage.

Inventive Principle:
Principle #5Merging (Combining)

2Reliability

If separate face detector and person detector are used, then detection capabilities are enhanced, but device complexity increases

Engineering Contradiction:
Improvedetection capabilitiesVSAvoiddetector architecture
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges face detection and person detection into a single detector architecture. The unified detector processes input images through one neural network model that simultaneously generates face detection outputs and person detection outputs, simplifying the overall system architecture while preserving the detection capabilities of both functions.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The detector is designed as a universal multi-functional model that can perform both face detection and person detection tasks. The single detector adapts to different detection objectives by producing different types of outputs (face bounding boxes and person bounding boxes) from the same input, reducing the need for multiple specialized detectors.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Reliability

If separate face detector and person detector are used, then detection scope is expanded, but information consistency deteriorates due to conflicting detections

Engineering Contradiction:
Improvedetection scopeVSAvoidinformation consistency
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

By merging face and person detection into a single detector, the patent ensures that both detection results are derived from the same feature extraction and processing pipeline. This unified approach eliminates conflicts between separate detectors since they operate independently, and enables direct association between face detections and person detections from the same processing context, improving information consistency.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentUS20240321002A1Whole Person Association with Face Hallucination and Screening
Publication Date: 2024.09.26 GOOGLE LLC
  • US20240321002A1 patent drawing
  • US20240321002A1 patent drawing
  • US20240321002A1 patent drawing

AI summary

Example aspects of the present disclosure are directed to computing systems and methods that perform whole person association with face screening and/or face hallucination. In particular, one aspect of the present disclosure is directed to a multi-headed person and face detection model that performs both face and person detection in one model. Each of the face and person detection can find landmarks or other pose information and also a confidence score. The pose information for the face and person detections can be used to select certain face and person detections to associate together as a whole person detection, which can be referred to as an “appearance.”