Image Object Counting via Dynamic Likelihood Thresholding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis systems face challenges in accurately detecting and counting objects, such as human bodies or heads, due to the inclusion of low-certainty likelihoods in the counting process, which can lead to incorrect detections or missed counts.
Innovation Solution
An information processing apparatus that acquires likelihood information from partial regions of an image, adjusts the threshold for counting targets by excluding likelihoods below a predetermined value, and estimates the number of objects by focusing on higher likelihoods, thereby improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a threshold is applied to exclude low-certainty likelihoods from counting, then detection accuracy is improved, but the complexity of the detection process increases
Solution Approach 1:
The image is divided into multiple partial regions, and the likelihood thresholding process is applied to each partial region independently. This segmentation allows the complex detection process to be broken down into manageable sub-tasks, improving detection accuracy for each region while making the overall process more systematic and controllable
Solution Approach 2:
Different likelihood thresholds are applied to different partial regions based on their specific characteristics. Instead of using a uniform threshold across the entire image, the system adjusts thresholds locally for each partial region, optimizing detection accuracy for regions with varying object densities and characteristics
2Device complexity
If likelihood information from all regions is included in counting, then the counting process is simplified, but incorrect detections increase
Solution Approach 1:
The system extracts and excludes likelihood information that falls below the threshold from the counting process. By removing unreliable likelihood data from consideration, the system maintains a simplified counting procedure while ensuring that only high-certainty detections contribute to the final count, thereby improving detection reliability
Solution Approach 2:
The system changes the parameter of likelihood threshold dynamically based on partial region characteristics. By adjusting the threshold parameter for different regions, the system filters out unreliable detections in a controlled manner, maintaining counting simplicity while improving overall detection reliability through parameter optimization
Data Source
AI summary
An information processing apparatus acquires, from each of a plurality of partial regions obtained by dividing an input image, likelihood information indicating a likelihood indicating certainty of presence of a particular object, based on the likelihood information, determines a region where the likelihood is greater than or equal to a first predetermined value among the plurality of partial regions, as a region where a threshold is to be adjusted to be lower, and estimates a number of particular objects by counting the likelihood with respect to each of the partial regions by excluding a likelihood less than the threshold among the likelihoods included in the partial regions from counting targets.