Face Detection and Tracking for Embedded Vision Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face detection systems on embedded vision systems face challenges in achieving real-time performance due to high complexity and limited computational resources, particularly with the MTCNN framework, and often result in tracking near-duplicate faces, wasting resources and bandwidth.
Innovation Solution
A process for real-time face detection, tracking, and pose selection that includes receiving a best-pose-face image, extracting image features, computing similarity values with stored features, and selectively transmitting unique images to a server, while using a CNN-based face detection module trained on grayscale images for efficient face detection on resource-limited systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If MTCNN framework is used for face detection, then face detection accuracy is improved, but system complexity and computational resource consumption increase
Solution Approach 1:
The MTCNN framework is segmented into multiple cascaded stages (first stage for candidate generation, second stage for refinement, third stage for final detection). Each stage processes only relevant regions, dividing the complex detection task into manageable segments that reduce overall computational burden while maintaining high accuracy
Solution Approach 2:
The system applies face detection at selective stages rather than continuously processing every frame. The first stage performs coarse detection on all frames, while subsequent refined detection is applied only to regions of interest or when confidence thresholds are not met, reducing redundant computations
2Measurement precision
If MTCNN framework is used for face detection, then face detection accuracy is improved, but processing speed decreases
Solution Approach 1:
The detection process is divided into sequential stages with decreasing computational intensity. The first stage uses simplified models for rapid candidate generation, the second stage refines only promising candidates, and the third stage performs detailed analysis on final candidates, achieving speed-accuracy tradeoff
Solution Approach 2:
The system performs full MTCNN detection periodically or at key moments rather than continuously. Between full detection cycles, simpler tracking or detection methods are used, creating a periodic pattern of intensive computation followed by lighter processing, which improves average processing speed
3Quantity of substance
If all detected face images are transmitted to server, then complete face data is available, but network bandwidth and storage resources are wasted
Solution Approach 1:
The system performs preliminary duplicate detection and pose quality assessment on the embedded device before transmission. By pre-processing and filtering face images locally, the system identifies and eliminates duplicates and low-quality images, transmitting only unique, high-value images to the server
Solution Approach 2:
The embedded vision system acts as an intermediary between the camera and the server, performing intelligent filtering and selection of face images. This intermediary function prevents unnecessary data transmission by blocking duplicates and low-quality images before they consume network bandwidth
4Measurement precision
If color images are used for training, then detection accuracy on color images is improved, but performance on grayscale inputs deteriorates
Solution Approach 1:
The CNN model is designed with universal feature extraction capabilities that work across different image types. By training with grayscale images in addition to color images, the model learns both color-specific and intensity-based features, making it adaptable to multiple input types including color, grayscale, and mixed scenarios
Data Source
AI summary
Embodiments described herein provide various examples of a real-time face-detection, face-tracking, and face-pose-selection subsystem within an embedded vision system. In one aspect, a process for identifying near-duplicate-face images using this subsystem is disclosed. This process includes the steps of: receiving a determined best-pose-face image associated with a tracked face when the tracked face is determined to be lost; extracting an image feature from the best-pose-face image; computing a set of similarity values between the extracted image feature and each of a set of stored image features in a feature buffer, wherein the set of stored image features are extracted from a set of previously transmitted best-pose-face images; determining if any of the computed similarity values is above a predetermined threshold; and if no computed similarity value is above the predetermined threshold, transmitting the best-pose-face image to a server and storing the extracted image feature into the feature buffer.