Animal Detection via Keypoint Probability and Affinity Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current animal detection systems in breeding factories face challenges in accurately counting animals, especially when they are crowded together, due to issues with overlapping bodies and slow detection speeds, leading to increased labor costs and detection delays.
Innovation Solution
A system and method that uses an animal detection model to generate probability maps and affinity field maps for key points, determining connection graphs to outline animal presence, allowing for real-time detection of animals with high accuracy, even when they are closely gathered.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual counting methods are used to count animals in breeding factories, then detection accuracy can be maintained through human observation, but labor costs increase and detection efficiency decreases
Solution Approach 1:
The patent replaces manual mechanical counting with an automated computer vision system that uses deep learning models to detect and count animals. The system processes images through neural networks to generate probability maps and affinity field maps, automatically identifying animal locations and counts without human intervention, thereby eliminating labor costs while improving detection efficiency.
2Measurement precision
If traditional object detection models are used to detect crowded animals, then the system structure remains simple, but detection accuracy decreases when animals are closely gathered
Solution Approach 1:
The patent segments the animal detection problem into multiple specialized components: a keypoint detection model that identifies specific anatomical points on each animal, and an affinity field model that determines relationships between keypoints to distinguish individual animals. This segmentation allows the system to handle crowded scenarios by focusing on discrete keypoint locations rather than attempting to detect entire animal bodies, thereby improving accuracy while managing model complexity through modular architecture.
Solution Approach 2:
The patent transitions from traditional 2D bounding box detection to a multi-dimensional approach by generating probability maps and affinity field maps that add depth information. The system uses three-dimensional spatial relationships between keypoints to distinguish overlapping animals, effectively adding a depth dimension to the detection problem that enables accurate counting of crowded animals that would be indistinguishable in standard 2D images.
3Speed
If fast detection algorithms are used to achieve real-time counting, then detection speed increases, but detection accuracy decreases leading to more miss detections
Solution Approach 1:
The patent performs preliminary processing by generating probability maps that pre-identify likely keypoint locations before final animal counting occurs. The affinity field maps are computed in advance to establish spatial relationships between keypoints, allowing the system to quickly resolve individual animal identities during the final counting stage. This preliminary action separates computationally intensive processing from the final detection output, enabling real-time performance with high accuracy.
Data Source
AI summary
A system and a method for detecting animals in a region of interest are disclosed. An image that captures a scene in the region of interest is received. The image is fed to an animal detection model to produce a group of probability maps for a group of key points and a group of affinity field maps for a group of key point sets. One or more connection graphs are determined based on the group of probability maps and the group of affinity field maps. Each connection graph outlines a presence of an animal in the image. One or more animals present in the region of interest are detected based on the one or more connection graphs.