Face Tracking Using Color Histograms to Reduce Computational Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face tracking in multimedia content is prone to inaccuracies due to the limitations of pattern recognition-based detectors, leading to missed faces and increased computational intensity when re-detecting faces in subsequent frames.
Innovation Solution
A method involving color-tracking using luminance projection and color histogram distances to re-detect faces in subsequent frames, reducing computational load by shifting detection window starting points and dividing face detection across multiple scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pattern recognition based detector is used for face detection in every frame, then detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the face tracking process into two distinct phases: (1) initial face detection using pattern recognition based detector, and (2) subsequent face tracking using simpler color-based tracking algorithms. This segmentation allows the computationally intensive pattern recognition detector to be used only when necessary (initial detection), while simpler methods handle routine tracking, thereby reducing overall computational complexity while maintaining detection accuracy.
Solution Approach 2:
The patent performs preliminary face detection using the pattern recognition based detector to establish initial face location and color characteristics. This preliminary action stores reference color information that enables subsequent tracking to rely on simpler color comparison methods rather than repeating the full pattern recognition process, thus reducing computational complexity in subsequent frames while maintaining tracking accuracy.
2Reliability
If pattern recognition based detector is re-used after losing track, then face detection accuracy is maintained, but number of computations per frame increases
Solution Approach 1:
The patent implements a feedback mechanism where the system continuously monitors tracking status and automatically switches between tracking modes. When face tracking is successful, the system uses efficient color-based tracking. When tracking is lost, the system triggers a re-detection phase using pattern recognition detector, then returns to color-based tracking. This feedback-controlled mode switching maintains tracking robustness while minimizing the frequency of computationally intensive re-detections.
Solution Approach 2:
The patent dynamically adjusts the tracking algorithm based on real-time performance conditions. The system transitions between static color-based tracking (when conditions are favorable) and dynamic pattern recognition-based re-detection (when tracking is lost). This dynamic adaptation allows the system to maintain robustness by using the more reliable pattern recognition detector only when necessary, rather than continuously, thereby reducing average computations per frame.
3Reliability
If color-tracking is performed continuously, then face tracking robustness is improved, but computational load increases
Solution Approach 1:
The patent employs periodic color-based tracking updates combined with event-triggered pattern recognition re-detection. Instead of continuously executing the full pattern recognition algorithm, the system performs lighter color-based tracking periodically and only invokes the computationally expensive pattern recognition detector when tracking confidence drops below a threshold or when face loss is detected. This periodic action with conditional reinforcement maintains tracking robustness while significantly reducing average computational energy consumption compared to continuous full-scale detection.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Method, apparatus,and computer program product are provided. The method includes detecting a face portion in a frame of a plurality of frames of a multimedia content. The method further includes tracking the face portion in at least one subsequent frame of the frame. Acolor-tracking of the face portion is performed on losing a track of the face portion in the at least one subsequent frame. The color-tracking is performed for re-tracking the face portion in the at least one subsequent frame.