Parallel Face Detection in Video Streams

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video coding technologies face challenges in real-time face detection in high-definition video streams, particularly in complex scenes, due to high computation demands and latency issues, which affect the quality of video conferencing applications.

Innovation Solution

A distributed coding architecture using multiple interconnected processors that perform parallel processing of video streams, employing block-level features such as temporal difference and edge features, along with color information, to identify face regions and adapt quantization accordingly, thereby reducing latency and improving detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If distributed coding architecture with multiple processors is used, then processing speed and low latency are improved, but system complexity increases

Engineering Contradiction:
Improveprocessing speedVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The video processing system is divided into multiple independent coding processors, each handling specific portions of video frames in parallel. This segmentation enables simultaneous processing of different frame regions, achieving high-speed real-time face detection while distributing computational load across multiple units, thus improving productivity without overwhelming single-point complexity

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Each coding processor in the distributed architecture is designed with multi-functionality, capable of performing both video coding operations and face detection tasks. This universal design allows the same hardware resources to serve multiple purposes, reducing overall system complexity while maintaining high processing throughput through parallel operation

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

2Measurement precision

If face detection is performed on entire picture, then detection accuracy is improved, but processing latency increases

Engineering Contradiction:
Improvedetection accuracyVSAvoidprocessing latency
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The face detection process is segmented to operate on specific regions of interest rather than the entire picture. By identifying and processing only relevant frame portions in parallel across multiple coding processors, the system maintains detection accuracy while significantly reducing the computational time required, thus lowering processing latency

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Motion estimation and block-level feature analysis are performed preliminarily during the video coding process before face detection. This preliminary processing prepares data structures and identifies potential face regions in advance, enabling faster and more accurate face detection without requiring re-processing of the entire picture, thereby reducing latency

Inventive Principle:
Principle #10Preliminary action

3Loss of time

If block-based analysis with motion estimation is used, then computation time is reduced, but detection accuracy in complex backgrounds deteriorates

Engineering Contradiction:
Improvecomputation timeVSAvoiddetection accuracy
Core Design Contradiction:
Loss of timeVSMeasurement precision

Solution Approach 1:

Multiple detection cues including skin color information, motion estimation results, and block-level features are merged into a unified face detection framework. This combination allows the system to leverage the computational efficiency of block-based analysis while compensating for its accuracy limitations in complex backgrounds through additional color and motion information from multiple sources

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system implements feedback mechanisms where detection results from block-based analysis are used to guide subsequent refinement steps. Motion-identified regions and color-segmented areas provide feedback that adjusts detection parameters and focuses computational resources on promising regions, improving accuracy in complex backgrounds without substantially increasing overall computation time

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP2135457B1Real-time face detection
Publication Date: 2013.09.18 CISCO TECHNOLOGY INC
  • EP2135457B1 patent drawingFigure 1
  • EP2135457B1 patent drawingFigure 2A
  • EP2135457B1 patent drawingFigure 2B

AI summary

An apparatus, a method, and a computer-readable medium having instructions encoded thereon that when executed cause a method to be carried out. The method includes dividing at least a portion of a picture of a video stream into parts of blocks, and processing the parts in parallel by a plurality of interconnected processors. The processing of a respective part by its respective processor includes determining block- features, e.g., block level temporal features and/or block level edge features. Each processor also performs coding functions on its respective part of the picture. The method also includes block-level processing using the block-level features to determine which blocks in the picture are likely to be that of a face, the block-level processing being at the granularity of at least a block.