Print Email Facebook Twitter Mice Tracking Using Infrared Subcutaneous Implants for Error Detection Title Mice Tracking Using Infrared Subcutaneous Implants for Error Detection Author Guinea Perez, Javier (TU Delft Electrical Engineering, Mathematics and Computer Science) Contributor van der Veen, A.J. (mentor) Nassirinia, F. (mentor) Marroquim, Ricardo (mentor) Degree granting institution Delft University of Technology Programme Computer Science | Software Technology Date 2021-04-06 Abstract The objective of this thesis was to develop a rodent tracker using the FlashTrack implants, and allow for tracking data to be used for behavioural research. The task was split into three main problems: detection, tracking and error detection. The detection was solved with basic image processing. The first step was background removal, using median filtering. Later, blob detection after some processing was done. Detections were differentiated into cases where mice are together in contact events and where they are alone. Bounding boxes were generated for the contours of lone mice, while the distance transform was employed to detect the joint ones. To track the moving targets, Kalman filtering was used on the bounding boxes of the detector. This approach was based on the Simple Online Realtime Tracking framework, adapting it to the particularities of mice. Other approaches were tried, but SORT was the chosen one. The infrared subcutaneous implants of FlashTrack allow for identity verification through code detections. To process the codes of each track, a Gaussian Mixture Model is trained to be the classifier of the detections. A track handling module was built to monitor the estimated tracks and code detections, and verify correct assignments. Detected erroneous tracks were discarded. Synthetic data was used to evaluate the tool. Artificial datasets were developed in Blender. Common metrics for evaluation of multiple object trackers were gathered and discussed, as well as a comparison with one of the state of the art animal trackers. Subject TrackingNeuroscienceRodentKalmanError DetectionMultiple Object Tracking To reference this document use: http://resolver.tudelft.nl/uuid:1d895f4c-76dc-4da6-b306-b9f622237658 Embargo date 2023-04-15 Part of collection Student theses Document type master thesis Rights © 2021 Javier Guinea Perez Files PDF Javier_Guinea_Thesis.pdf 10.41 MB Close viewer /islandora/object/uuid:1d895f4c-76dc-4da6-b306-b9f622237658/datastream/OBJ/view