Deep Neural Network Image Mask Generation for Moving Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object segmentation methods are either inaccurate or too slow, and they struggle with segmenting moving or deforming objects due to errors in static background determination and unreliable depth information, leading to segmentation errors and inefficiencies.
Innovation Solution
An electronic device and method using a deep neural network (DNN) for image mask generation, which applies a trained convolutional neural network (CNN) to segment objects in real-time by generating a pilot mask, extracting a region of interest, resizing it to a threshold scaling size, and reapplying the DNN to produce a final mask, ensuring precise segmentation with minimal error and improved processing speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional static background subtraction method is used, then processing speed is fast, but segmentation accuracy deteriorates due to errors in determining static background
Solution Approach 1:
The system performs preliminary actions by capturing multiple sequential images of the background scene before the actual segmentation task, uses these pre-captured images to generate a static background model, and only then proceeds to subtract this model from current frames. This preliminary background acquisition phase enables accurate segmentation without slowing down the real-time processing of subsequent frames.
Solution Approach 2:
The method segments the background into multiple sequential images captured at different times, processes each image to identify and remove moving objects, generates a cleaned background model from these segmented images, and then uses this model for foreground extraction. This segmentation approach allows the system to handle complex scenes with moving objects during background capture without compromising final segmentation accuracy.
2Measurement precision
If depth information is used for foreground identification, then segmentation accuracy improves, but reliability deteriorates due to noise in depth map sensors
Solution Approach 1:
The system introduces an intermediary approach by using multiple sequential background images as a mediator between the captured frame and the final segmentation result. Instead of directly relying on noisy depth information, the system uses the statistical consistency across multiple background frames to create a robust model that indirectly identifies foreground objects, filtering out noise from depth sensors.
Solution Approach 2:
The method implements feedback by iteratively refining the background model through multiple sequential image processing cycles. Each cycle processes background images, identifies and removes moving objects, and updates the background model. This feedback mechanism allows the system to correct errors in depth information by cross-validating against multiple observed background states.
3Measurement precision
If shape based object segmentation is used, then segmentation accuracy improves, but processing speed deteriorates
Solution Approach 1:
The system performs preliminary background modeling using multiple sequential images captured and processed in advance. By pre-computing a robust background model that accounts for various scene conditions and moving objects, the system eliminates the need for computationally intensive shape-based analysis during real-time processing, thereby maintaining high processing speed while achieving accurate segmentation.
Data Source
AI summary
An electronic device and method for image mask generation using a deep neural network is provided. The electronic device stores an input image that includes an object of interest generates a pilot mask for the input image by application of a trained deep neural network (DNN) on the input image. The electronic device extracts, from the input image, a region of interest that includes the object of interest, based on the generated pilot mask and resizes the extracted region of interest such that a size of the resized region of interest equals a threshold scaling size. The electronic device further generates a final mask by reapplication of the trained DNN on the resized region of interest.