Object Density Map Generation Using Gated Shortcut Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object density estimation systems face scalability issues due to their reliance on video streams and require calibration, limiting their deployment in various scenarios and increasing computational load.
Innovation Solution
A method using a deep neural network with compression and decompression layers connected by gated shortcuts to generate object density maps from still images, allowing for object density monitoring without prior calibration and enabling operation on independent images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If video streams are used for object density estimation, then object counting can be performed, but computational load increases and scalability decreases
Solution Approach 1:
The patent extracts the essential information needed for density estimation from video streams by processing individual frames independently. Instead of analyzing temporal sequences, the method extracts density information from single images, removing the computational burden of video processing while maintaining counting accuracy.
Solution Approach 2:
The patent segments the object counting problem into independent frame processing tasks. Each image frame is processed separately through the neural network, allowing parallel processing and eliminating the need for continuous video stream analysis, thus reducing computational load while maintaining measurement precision.
2Measurement precision
If calibration is performed for object density estimation, then measurement accuracy improves, but system complexity and deployment difficulty increase
Solution Approach 1:
The neural network performs self-calibration by learning to estimate object density directly from image content without external calibration data. The system serves itself by automatically adapting to different scenes and object types through the annotated training dataset, eliminating the need for manual calibration procedures.
Solution Approach 2:
The patent changes the approach from fixed calibration parameters to learned parameters through neural network training. The system adapts its internal parameters automatically during training with annotated datasets, replacing manual calibration with automated parameter optimization that works across diverse deployment scenarios.
3Ease of operation
If object detection is used for counting, then individual objects can be localized, but accuracy deteriorates as object density increases
Solution Approach 1:
Instead of detecting individual objects and then counting them, the patent inverts the approach by directly estimating object density from image features. Rather than localizing each object to count, the system predicts density maps that indicate concentration regions, which is more effective for dense scenes where individual detection becomes unreliable.
Data Source
AI summary
A method for object density monitoring includes receiving, by a processing server, an input image captured by an image sensor. The method further includes providing an annotated dataset with a target object to be identified in the input image, and providing, by the processing server as output, an object density map generated from the input image. The processing server provides the object density map by using a deep neural network having one or more pairs of a compression layer and a decompression layer connected by gated shortcuts.