Human Image Tracking Using Depth Sensor and Graph Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Surveillance systems face challenges in detection precision and instantaneity, particularly with side view systems due to occlusion and poor performance in dim environments, and top view systems mistaking floor lamps for human heads.

Innovation Solution

A human image detection and tracking system using a depth image sensor and graph-based segment processing to identify human regions, with a hemisphere model for head detection and depth distribution analysis for tracking, improving detection precision and processing speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If side view type surveillance systems are used to detect and track human images, then the system can operate in various environments, but occlusion of tracked targets occurs making it difficult to identify missing parts and the system does not operate well in dim environments

Engineering Contradiction:
Improveenvironmental adaptabilityVSAvoiddetection precision
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent transitions from 2D color image surveillance to 3D depth image surveillance. By capturing depth information as an additional dimension, the system can identify targets even when occluded in 2D views, and can operate effectively in dim environments where color cameras fail. The depth data provides spatial information that resolves occlusion problems inherent in side-view 2D systems.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If top view type surveillance systems are used to distinguish human images by extracting depth data, then human images can be distinguished, but floor lamps are often mistaken for a person's head

Engineering Contradiction:
Improvedetection precisionVSAvoiddetection reliability
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent applies local quality analysis by examining the spatial distribution characteristics of depth data in different regions. Specifically, it analyzes the vertical depth distribution pattern - human heads exhibit a characteristic pattern where the top of the head is closer to the camera than the body, creating a specific depth gradient. Floor lamps lack this pattern, allowing the system to distinguish between them despite both appearing as vertical objects in top-view depth data.

Inventive Principle:
Principle #3Local quality

3Productivity

If conventional surveillance systems are used for monitoring, then basic video recording can be achieved, but detection precision and instantaneity remain insufficient for identifying suspicious people

Engineering Contradiction:
Improveprocessing speedVSAvoiddetection precision
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent extracts and utilizes the depth information channel separately from the color image channel. By focusing on depth data for detection and tracking purposes, the system achieves both high speed (through efficient depth data processing) and high precision (through reliable 3D spatial information). This extraction of the depth channel enables real-time detection with improved accuracy compared to conventional 2D video monitoring.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentUS9317765B2Human image tracking system, and human image detection and human image tracking methods thereof
Publication Date: 2016.04.19 NAT TAIWAN UNIV
  • US9317765B2 patent drawing
  • US9317765B2 patent drawing
  • US9317765B2 patent drawing

AI summary

A human image detection and tracking systems and methods are disclosed. A human image detection method comprises receiving a depth image data from a depth image sensor by an image processing unit, removing a background image of the depth image sensor and outputting a foreground image by the image processing unit, receiving the foreground image and operating a graph-based segment on the foreground image to obtain a plurality of graph blocks by a human image detection unit, determining whether a potential human region exists in the graph blocks, determining whether the potential human region is a potential human head region, determining whether the potential human head region is a real human head region, and regarding the position of the real human head region is the human image position by the human image detection unit if the potential human head region is the real human head region.