Improving the State Estimation Accuracy of Real-Time Vision Based Multiple Target Tracking Algorithms with Unequal Dimension Interactive Multiple Model Estimator

Date

2023-6-22

Author

Kurt, Yağız

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

103
views

72
downloads

Cite This

This study presents methodology for improving the vision-based multiple object motion estimation performance for the maneuvering objects in case of occlusion. Present deep learning-based visual MOT (Multiple Object Tracking) algorithms cannot track the objects when an occlusion exists even for a short period of time. When the measurement is not available, the prediction model propagates the Kalman Filter’s motion model, or the tracker maintains the previous estimation until a measurement is received. In either case, during these periods, the object may maneuver or accelerate/decelerate so that the tracker cannot assign the tracklets successfully when the measurement is arrived. Using Unequal dimension Interactive Multiple Models (UDIMM) approach, track loss due to the ID switch is highly reduced compared to the simple constant velocity motion model used in the StrongSort multi-object tracker algorithm. For the concerns on real time computation on onboard aerial vehicles, lightweight object detector and MOT algorithms, YOLOv5 and StrongSort are implemented, respectively. Estimation and performance results of different filters are compared on simulation environment.

Subject Keywords

interactive multiple model filter, vision based multiple object tracking, estimation, deep learning

URI

https://hdl.handle.net/11511/104774

Collections

Graduate School of Natural and Applied Sciences, Thesis