Near-Online Multi-target Tracking with Aggregated Local Flow Descriptor

By Wongun Choi. IEEE International Conference on Computer Vision, 2015

430 citations recorded by the source corpus.

In this paper, we tackle two key aspects of multiple target tracking problem: 1) designing an accurate affinity measure to associate detections and 2) implementing an efficient and accurate (near) online multiple target tracking algorithm. As for the first contribution, we introduce a novel Aggregated Local Flow Descriptor (ALFD) that encodes the relative motion pattern between a pair of temporally distant detections using long term interest point trajectories (IPTs). Leveraging on the IPTs, the ALFD provides a robust affinity measure for estimating the likelihood of matching detections regardless of the application scenarios. As for another contribution, we present a Near-Online Multi-target Tracking (NOMT) algorithm. The tracking problem is formulated as a data-association between targets and detections in a temporal window, that is performed repeatedly at every frame. While being efficient, NOMT achieves robustness via integrating multiple cues including ALFD metric, target dynamics, appearance similarity, and long term trajectory regularization into the model. Our ablative analysis verifies the superiority of the ALFD metric over the other conventional affinity metrics. We run a comprehensive experimental evaluation on two challenging tracking datasets, KITTI [16] and MOT [2] datasets. The NOMT method combined with ALFD metric achieves the best accuracy in both datasets with significant margins (about 10% higher MOTA) over the state-of-the-art.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Electrical and Electronic Engineering

View Paper: https://www.semanticscholar.org/paper/db9177836333818f9b4c9cdd6c74cffaa608a1f8

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

In Electrical and Electronic Engineering

Citations in the corpus

VirtualWorlds as Proxy for Multi-object Tracking Analysis (2016, methodology): “Thanks to the recent progress on object detection, association-based tracking-by-detection in monocular video streams is particularly successful and widely used for MOT [29, 30, 31, 32, 33, 34, 35, 36, 37, 38] (see [39] for a recent review).”
Simple online and realtime tracking with a deep association metric (2017, neutral): “2% 1394 1318 9855 57257 35 Hz NOMTwSDP16 [19] BATCH 62.”
Spatial-Temporal Relation Networks for Multi-Object Tracking (2019, neutral): “Offline NOMT [11] 46.”
FairMOT: On the Fairness of Detection and Re-identification in Multiple Object Tracking (2020, methodology): “Multi-object tracking can be classified into online methods [1], [24], [25], [26], [27] and batch methods [28], [29], [30], [31], [32], [33] based on whether they rely on future frames.”
PoseTrack: A Benchmark for Human Pose Estimation and Tracking (2017, methodology): “Given the joint localization threshold for each person, we compute two sets of evaluation metrics, one which is commonly used for evaluating multi-person pose estimation [35], and one from the multi-target tracking literature [50, 8, 30] to evaluate multiperson pose tracking.”
Opening up Open World Tracking (2021, background): “Early effort focused on developing strong data association techniques [11,39,41,50,63,65] and hand-crafting appearance [23, 49, 52] and motion cues [15, 38].”
Learning a Neural Solver for Multiple Object Tracking (2019, background): “A recent trend is to design ever more complex models which include other vision input such as reconstruction for multi-camera sequences [49, 79], activity recognition [17], segmentation [55], keypoint trajectories [15] or joint detection [75].”
Joint Graph Decomposition & Node Labeling: Problem, Algorithms, Applications (2016, methodology): “Here, we apply to the joint NL - LMP the algorithms KLj/r and KLj ∗ r and compare their output to that of [36] and of other top-performing algorithms [8, 12, 19].”
MOTChallenge: A Benchmark for Single-Camera Multiple Target Tracking (2020, methodology): “91 NOMT [Choi, 2015] 33.”
Learning by Tracking: Siamese CNN for Robust Target Association (2016, methodology): “To address these issues, research in this area has produced more and more complex models: global optimization methods based on network flow [4, 64], minimum cliques [61] or discrete-continuous CRF inference [1]; models of pedestrian interaction with social motion models [35, 44]; integration of addi”
3D Multi-Object Tracking: A Baseline and New Evaluation Metrics (2019, methodology): “Modern methods often use the motion model [19], [25], [26] and the appearance feature [25], [27], [28].”
Online Multi-Object Tracking with Dual Matching Attention Networks (2018, neutral): “1 NOMT [9] 46.”
Tracking the Untrackable: Learning to Track Multiple Cues with Long-Term Dependencies (2017, neutral): “NOMT [9] Offline 46.”
Joint Graph Decomposition and Node Labeling by Local Search (2016, methodology): “Here, we apply to the joint NL-LMP the algorithms KLj-r and KLj+r and compare their output to [ 26] and to other top-performing algorithms [8, 13, 4, 27].”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, methodology): “A similar performance achieves the near-online multi-target tracking algorithm presented by Choi (2015) formulated as a global data association problem.”
Chained-Tracker: Chaining Paired Attentive Regression Results for End-to-End Joint Multiple-Object Detection and Tracking (2020, neutral): “Offline NOMT [34] 62.”