POI: Multiple Object Tracking with High Performance Detection and Appearance Feature

By Fisher Yu, Wenbo Li, Quanquan Li, Yu Liu, Xiaohua Shi, Junjie Yan. ECCV Workshops, 2016

475 citations recorded by the source corpus.

Detection and learning based appearance feature play the central role in data association based multiple object tracking (MOT), but most recent MOT works usually ignore them and only focus on the hand-crafted feature and association algorithms. In this paper, we explore the high-performance detection and deep learning based appearance feature, and show that they lead to significantly better MOT results in both online and offline setting. We make our detection and appearance feature publicly available (https://drive.google.com/open?id=0B5ACiy41McAHMjczS2p0dFg3emM). In the following part, we first summarize the detection and appearance feature, and then introduce our tracker named Person of Interest (POI), which has both online and offline version (We use POI to denote our online tracker and KDNT to denote our offline tracker in submission.).

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/beb0239feac388e4ee04492159a45f7e2c71e1e3

How researchers see this paper

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

In Computer Vision and Pattern Recognition

Citations in the corpus

SG-Net: Spatial Granularity Network for One-Stage Video Instance Segmentation (2021, methodology): “Some methods use motion priors, such as Kalman Filter [12, 45] or optical flow [41] to associate the detected bounding boxes.”
Memory Based Online Learning of Deep Representations from Video Streams (2017, background): “Few exceptions can operate online and use deep features [41, 42, 43, 44] but they still assume continuous motion and do not update the appearance.”
TransTrack: Multiple-Object Tracking with Transformer (2020, methodology): “These approaches have achieved the best performance on the public dataset [42].”
Deep learning in video multi-object tracking: A survey (2019, methodology): “In fact, the algorithm presented in [38], that uses a modified Faster R-CNN, has held its top ranking position among online methods on MOT16 for 3 years, and many of the other top-performing MOT16 algorithms have employed the same detections.”
FairMOT: On the Fairness of Detection and Re-identification in Multiple Object Tracking (2020, methodology): “As a result, these approaches (Yu et al. 2016; Henschel et al. 2019) have achieved the best performance on the public datasets.”
Features for Multi-target Multi-camera Tracking and Re-identification (2018, methodology): “This idea has been shown to work for person detection [17], and implicitly pursued in single-camera trackers [66, 67, 73] and Re-ID methods [35, 82, 83] that improve features to increase accuracy.”
Joint Monocular 3D Vehicle Detection and Tracking (2018, methodology): “We utilize a mixture of those factors as the affinity across frames, similar to the design of POI [49].”
Quasi-Dense Similarity Learning for Multiple Object Tracking (2020, methodology): “Some methods use motion priors, such as Kalman Filter [4, 52], optical flow [48], and displacement regression [10, 16], to ensure accurate distance estimations.”
Evaluating Merging Strategies for Sampling-based Uncertainty Techniques in Object Detection (2018, methodology): “models will be deployed under closed-set conditions [2], [16], i.”
Adaptive RNN Tree for Large-Scale Human Action Recognition (2017, methodology): “Last, with the assistance of recent tracking algorithms [4, 15, 30, 33] and the mobile computing techniques [12, 13], we hope to apply our method to more complex action recognition in real life.”
Simple online and realtime tracking with a deep association metric (2017, methodology): “Methods marked with use detections provided by [16].”
Multi-Target Multi-Camera Tracking by Tracklet-to-Target Assignment (2020, background): “Due to the impressive progress of object detection techniques [51]–[53], tracking-by-detection [16], [54]–[57] has become the mainstream approach for multi-target tracking in recent years.”
Tracking Objects as Points (2020, methodology): “We believe that there is an exciting avenue for future work in combining local trackers (such as our work) with long-range models (such as SORT [2], LMP [41], and other ReID-based trackers [50, 52]).”
StrongSORT: Make DeepSORT Great Again (2022, methodology): “DeepSORT uses the optimized Faster R-CNN [39] presented in [60] as the detector and trains a simple CNN as the embedding model.”
Track to Detect and Segment: An Online Multi-Object Tracker (2021, unknown): “2% 9,855 57,257 1,318 1,394 27+D POI[59] ECCVW 2016 66.”
BoT-SORT: Robust Associations Multi-Pedestrian Tracking (2022, methodology): “Currently, tracking-by-detection has become the most effective paradigm for the MOT task [54, 3, 48, 4, 58].”