Rotated Bounding Boxes for Animal Mounting Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing techniques struggle to accurately detect mounting behavior in multiple dynamically moving animals, particularly in crowded environments, due to overestimation of object areas and misrecognition of motions, which affects detection accuracy.
Innovation Solution
A specialized animal detection method that creates bounding boxes aligned with the rotational direction of animal objects, using animal detection models trained on learning data, including bounding box coordinates, rotational angles, and activity information from motion history images, to accurately identify mounting behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If axis-aligned bounding boxes are used for object detection, then detection process is simple, but detection accuracy decreases when objects are aligned diagonally or crowded
Solution Approach 1:
The patent applies asymmetry by transitioning from axis-aligned (symmetric) bounding boxes to rotated bounding boxes that can be oriented at any angle. This allows the detection framework to adapt to objects positioned diagonally or in various orientations, improving detection accuracy while maintaining computational efficiency through the use of rotation matrices and transformed coordinate systems.
Solution Approach 2:
The patent implements dynamics by making the bounding box orientation adjustable and adaptive rather than fixed. The system dynamically determines the optimal rotation angle for each detected object based on its actual orientation in the image, allowing the bounding box to rotate and align with the object's direction, thereby improving precision in crowded and diagonally-aligned scenarios.
2Ease of operation
If conventional object detection methods are used, then processing is straightforward, but mounting behavior detection accuracy decreases in crowded environments
Solution Approach 1:
The patent applies segmentation by first detecting individual animal objects and their bounding boxes, then separately analyzing the spatial relationships and orientations between pairs of objects. This two-stage approach segments the complex mounting behavior detection task into object detection and behavior classification components, maintaining operational simplicity while improving accuracy through focused analysis of relevant features.
Solution Approach 2:
The patent introduces another dimension by adding rotational angle information to the traditional bounding box parameters. Instead of only considering axis-aligned positions, the system incorporates the orientation dimension, allowing it to distinguish mounting behavior based on the relative angles between animals, thereby improving detection accuracy in crowded environments without significantly complicating the processing workflow.
3Device complexity
If motion history image (MHI) alone is used for activity detection, then motion detection is simple, but accuracy decreases when many animals are present
Solution Approach 1:
The patent merges multiple detection approaches by combining motion history image (MHI) analysis with rotated bounding box orientation analysis. The system integrates temporal motion information from MHI with spatial orientation information from rotated bounding boxes, creating a more robust detection framework that improves accuracy in crowded environments while maintaining reasonable complexity through the synergistic combination of these two methods.
Data Source
AI summary
An apparatus for detecting mounting behavior of an animal object includes: a memory that stores a program; and a processor that executes the program. The program extracts animal detection information about an animal object detected from the image by inputting the received image into an animal detection model. Also, the program extracts bounding boxes of which a distance between coordinates of central points is smaller than a first set value, bounding boxes of which a difference in rotational angle is smaller than a second set value, and bounding boxes of which a difference between a vector connecting the central points of the extracted bounding boxes and an orientation of each bounding box is smaller than a third set value. If activity information of the animal object is extracted based on an MHI of the image, it is determined that mounting behavior occurs.