Tracking objects with several cameras using deep learning on Raspberry Pi

Details

Supervisors

Benoît Macq

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en sciences informatiques, à finalité spécialisée

Abstract

This thesis is interested in a multi-agent distributed tracking system that robustly tracks objects in the presence of obstructions. As part of the multi-agent network, the object detection approach is developed on many single-board computers, RaspberryPi. The existence of multiple cameras helps the central unit to compensate for occlusions by providing a variety of viewpoints on the objects. The tracking is done on the central unit once the detection data from multiple cameras is gathered via the application programming interface. Because camera observations are noisy, the Kalman filter with tracking is used to investigate the problem of noise reduction. The main focus of this thesis is to achieve consistent tracking results across camera views in the presence of obstructions and use multiple single-board computers to achieve a balance between accuracy and low cost.