Sort and Deep-SORT Based Multi-Object Tracking for Mobile Robotics: Evaluation with New Data Association Metrics

Multi-Object Tracking (MOT) techniques have been under continuous research and increasingly applied in a diverse range of tasks. One area particular concerns its application navigation tasks assistive mobile robots, with the aim to increase mobility autonomy people suffering from decay, or severe motor impairments, due muscular, neurological, osteoarticular decay. Therefore, this work, having view for an evaluation study two MOTs by detection algorithms, SORT Deep-SORT, is presented. To improve data association both methods, which are solved as linear assignment problem generated cost matrix, set new object tracking matrices based on intersection over union, Euclidean distances, bounding box metrics proposed. For MOT real-time pipeline, YOLOv3 used detect classify objects available images. In addition, perform proposed aiming at platforms, ISR dataset, represents conditions real robotic platforms may navigate, Experimental evaluations were also carried out MOT17 dataset. Promising results achieved matrices, showing improvement majority compared default matrix. promising frame rate values attained pipeline composed detector module.