Deep Learning Autofocus for Face Detection in Complex Lighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional camera autofocus systems face challenges in accurately focusing on a subject's face, especially in conditions like backlighting, small subject sizes, side-view faces, and moving subjects, leading to inefficiencies and the need for manual focus adjustments.
Innovation Solution
A deep-learning-based autofocus system utilizing a fully convolutional network algorithm processes image content to separate body and head areas, allowing for automatic or user-selected focus on a subject's face or head by computing distances and applying heat maps and masks to distinguish and isolate head regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional face detection methods are used for autofocus, then the system is simple and fast, but it fails in complex scenarios such as backlighting, small subjects, side-view faces, and moving subjects
Solution Approach 1:
The patent introduces an intermediary deep learning model that acts as a mediator between the image input and the autofocus mechanism. This model processes the image to generate segmented maps identifying body and head regions, enabling reliable autofocus in complex scenarios without directly modifying the core autofocus hardware
Solution Approach 2:
The patent applies segmentation by dividing the image into distinct regions (body areas and head areas) using a deep learning model. This segmentation allows the system to reliably identify and focus on the head region even when conventional face detection fails due to lighting, size, or orientation issues
2Reliability
If manual focus selection is used, then the photographer can select the desired autofocus point, but it requires waiting and reduces shooting speed
Solution Approach 1:
The patent implements self-service by enabling the camera system to automatically select and focus on the subject's head without requiring manual intervention. The deep learning model autonomously identifies the head region and guides the autofocus mechanism, eliminating the need for photographer waiting and manual adjustments while maintaining precise focus
3Measurement precision
If the subject is small in the image, then conventional face detection may fail, but increasing image processing complexity may slow down the system
Solution Approach 1:
The patent applies preliminary action by using the deep learning model to pre-process and segment the image into body and head regions before the autofocus operation. This preliminary segmentation identifies small subjects early, allowing the system to focus on the correct region without time-consuming trial-and-error adjustments during the actual autofocus process
Data Source
AI summary
To overcome several issues of autofocusing, a deep-learning-based autofocus system utilizes a subject's body to autofocus on the subject's face. The subject, including the subject's body and face/head, are determined utilizing image processing methods, and based on the detection, the subject's face/head is automatically focused on.