MOTChallenge: A Benchmark for Single-Camera Multiple Target Tracking

Abstract

Standardized benchmarks have been crucial in pushing the performance of computer vision algorithms, especially since the advent of deep learning. Although leaderboards should not be over-claimed, they often provide the most objective measure of performance and are therefore important guides for research.We present MOTChallenge, a benchmark for singlecamera Multiple Object Tracking (MOT) launched in late 2014, to collect existing and new data and create a framework for the standardized evaluation of multiple object tracking methods. The benchmark is focused on multiple people tracking, since pedestrians are by far the most studied object in the tracking community, with applications ranging from robot navigation to self-driving cars. This paper collects the first three releases of the benchmark: (i) MOT15, along with numerous state-of-the-art results that were submitted in the last years, (ii) MOT16, which contains new challenging videos, and (iii) MOT17, that extends MOT16 sequences with more precise labels and evaluates track-

Enhancing the association in multi‐object tracking via neighbor graph

“…7 We evaluate our approach on the most widely used MOT Challenge benchmark. 11 It achieves state-of-the-art overall performance on both MOT16 12 and MOT17 12 data sets,…”

“…We compare our method with the state-of-the-art trackers, including two graph-related methods (GSM 8 and GSDT 7 ), on the MOT Challenge platform. 11 The platform guarantees fairness and objectiveness with two measures: first, it provides unified data for training and testing; second, testing results must be uploaded to the platform for standardized evaluations, with the rule that any tracker has at most three chances for evaluating, which avoids overfitting and tricking. Results reported in Table 1 show that our tracker achieves the highest MOTA score on both of the MOT16 and MOT17 testing, which demonstrates the superiority of our method.…”

“…Note that here we do not set a distance threshold to filter out the unqualified ones from the selected neighbors because, in the scenario of vision MOT, such as MOT Challenge, 11 the video can be filmed by the first or bird view, static or moving camera, resulting in frequent and dramatic variations of objects' scale. Therefore, it is very difficult to set a proper distance threshold for various scenarios.…”

vmTracking: Virtual Markers Overcome Occlusion and Crowding in Multi-Animal Pose Tracking

“…Crowding is when the density of animals in a confined space is high, leading to errors in identifying and tracking specific animals. Despite efforts in research and development to resolve occlusion and crowding (Dendorfer et al, 2020; Han et al, 2023), these issues remain unresolved and hinder accurate multi-animal tracking and pose estimation.…”