Tracklets for Ambiguous Animal Identification in Multi-Housed Cages
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multihoused animal cages, existing identification methods such as RFID tags and visual observations are often unreliable due to animals burrowing or being in range of sensors, leading to ambiguity in animal identification, especially at night or with limited visibility.
Innovation Solution
The use of tracklets, which are defined as portions of an animal's path within a cage that begin and end at ambiguation events, allowing for unique identification of an animal by following its path and associating sensor data with that animal, even in situations where initial identification is ambiguous.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RFID tags or visual observation methods are used for animal identification, then identification can be performed, but reliability deteriorates when animals are burrowed under bedding or in range of sensors
Solution Approach 1:
The identification system divides the cage into multiple zones with different sensor types (visual sensors for open areas, RFID sensors for bedding areas). Each zone independently identifies animals using the most appropriate modality, and the system segments the identification problem into zone-specific subproblems that can be solved independently and then integrated.
Solution Approach 2:
The system introduces tracklets as intermediary data structures that bridge the gap between ambiguous sensor readings and definitive animal identification. Tracklets serve as temporary identifiers that accumulate evidence from multiple sensors over time, allowing the system to resolve ambiguities when animals are temporarily unidentifiable by any single sensor.
2Productivity
If multiple animals are placed in a single cage to improve housing efficiency, then productivity increases, but identification accuracy deteriorates due to animal proximity and ambiguity
Solution Approach 1:
The system dynamically adjusts identification strategies based on real-time animal positions and behaviors. When animals are close together and visual identification is ambiguous, the system switches to RFID-based identification or waits for animals to separate. Tracklets are dynamically created and merged based on movement patterns, allowing the system to handle varying levels of animal proximity without compromising identification accuracy.
Solution Approach 2:
The system performs preliminary identification attempts using multiple modalities before concluding that identification is ambiguous. Visual sensors continuously scan for identifying features, RFID sensors periodically check for tags, and the system prepares tracklets in advance to capture movement patterns. This preliminary action allows the system to resolve most identification cases without needing complex disambiguation logic.
3Difficulty of detecting and measuring
If visual sensors are used to track animal behavior, then behavior detection capability improves, but identification reliability deteriorates at night or with limited visibility
Solution Approach 1:
The system merges data from multiple independent identification modalities (visual sensors, RFID sensors, tracklet analysis) into a unified identification framework. Each modality contributes partial information, and the system combines these inputs to achieve reliable identification. For example, visual sensors detect behavior patterns while RFID sensors provide periodic identification confirmation, and tracklets link these data sources together.
Solution Approach 2:
The system uses periodic RFID scanning to complement continuous visual monitoring. RFID sensors periodically attempt to identify animals in the cage, providing identification confirmation even when visual identification is impossible due to lighting conditions or animal position. This periodic action ensures that identification reliability is maintained across all lighting conditions and animal states.
Data Source
AI summary
The field of this invention is recording motion of an animal, uniquely identifying the animal, and recording an activity of that animal. Motions of animals on a path when the animal is or can be uniquely identified are tracklets. Tracklets begin and end at ambiguation events, where these are defined as locations and times of an animal where it cannot be uniquely identified. A first animal may be uniquely identified by first identifying all other animals in the first animal's environment. Animal identification may be after the end of a tracklet. Embodiments use optical flow analysis from video of an animal's environment. Embodiments record an animal's activity on tracklets and then use that activity to measure animal health or use that activity as data for a study using animals.