Learning task-specific discriminative representations for multiple object tracking

Wu, Han; Nie, Jiahao; Zhu, Ziming; He, Zhiwei; Gao, Mingyu

doi:10.1007/s00521-022-08079-3

Cited by 10 publications

(9 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Wu et al [39] introduced a dual-path transformation network (DTN) for MOT which minimized the contradictions of optimization with the network through decoupling sharing features into the representation of specific tasks. The pyramid nonlocal network (PNN) was employed to improve the representation of specific Re-ID scaling data.…”

Section: Literature Surveymentioning

confidence: 99%

ResNet50-deep affinity network for object detection and tracking in videos

2024

IJATEE

View full text Add to dashboard Cite

Multiple-object tracking (MOT) plays a crucial role in addressing many fundamental challenges within the fields of computer vision and video analysis. The majority of MOT methods rely on two primary processes: object detection and data association. Initially, each video frame is analyzed to detect objects, followed by a subsequent step that establishes correlations among the detected objects across multiple frames to generate their tracks. However, the data association for tracking often relies on manually defined criteria such as motion, appearance, grouping, and spatial proximity, among others. In the study, the ResNet50-deep affinity network (DAN) was introduced, which had been designed for the detection and tracking of objects in videos, including those that appear and disappear between frames. The proposed method was evaluated using the widely recognized MOT17 dataset to address MOT challenges. During the preprocessing phase, photometric distortion correction, frame expansion, and cropping were performed. The ResNet50 model was utilized to extract features. The DAN was employed to identify object appearances in video frames and to calculate their cross-frame affinities (CFA). The approach was compared with existing research, including DAN, ByteTrack, the graph neural network for simultaneous detection and tracking (GSDT), the reptile search optimization algorithm with deep learning-based multiple object detection and tracking (RSOADL-MODT), the center graph network (CGTracker), the hybrid motion model, FlowNet2-deep learning, and the super chained tracker (SCT), to validate the efficiency of theResNet50-DAN method. The ResNet50-DAN method achieved superior results, with a multiple-object tracking accuracy (MOTA) of 84.2%, an F1 score for identification metrics (IDF1) of 80.3%, 10,352 false positives (FP), and 1,284 identity switches (ID-Sw). The ResNet50-DAN method demonstrated higher MOTA compared to the existing approaches, including DAN, ByteTrack, GSDT, RSOADL-MODT, CGTracker, the hybrid motion model, FlowNet2-DL, and SCT.

show abstract

Section: Literature Surveymentioning

confidence: 99%

ResNet50-deep affinity network for object detection and tracking in videos

2024

IJATEE

View full text Add to dashboard Cite

show abstract

“…Their lightweight version runs at 26.6 FPS. The YOLO series algorithm is generally around 50 FPS 20 . Wu et al took the approach of transforming shared features into detection-specific and ReID-specific representations.…”

Section: Related Workmentioning

confidence: 99%

A new detection algorithm for alien intrusion on highway

Guo

Lou

Chen

et al. 2023

Sci Rep

View full text Add to dashboard Cite

In recent years, highway accidents occur frequently, the main reason is that there is always foreign body invasion on the highway, which makes people unable to respond to emergencies in time. In order to reduce the occurrence of highway incidents, an object detection algorithm for highway intrusion was proposed in this paper. Firstly, a new feature extraction module was proposed to better preserve the main information. Secondly, a new feature fusion method was proposed to improve the accuracy of object detection. Finally, a lightweight method was proposed to reduce the computational complexity. We compare the algorithm in this paper with existing algorithms, the experimental results showed that: On the Visdrone dataset (small size targets), (a) the CS-YOLO was 3.6% more accurate than the YOLO v8. (b) The CS-YOLO was 1.2% more accurate than the YOLO v8 on the Tinypersons dataset (minimal size targets). (c) CS-YOLO was 1.4% more accurate than YOLO v8 on VOC2007 data set (normal size).

show abstract

“…In the field of MOT, tracking-by-detection (TBD) [ 3 ] stands out as the predominant paradigm, comprising three key sub-tasks. First, it detect objects from the current video frame.…”

Section: Introductionmentioning

confidence: 99%

“…Third, it associates the detected objects with those from the previous frame, relying on cues such as the similarity of the ReID features and intersection over union (IoU). Within the TBD framework, the architecture of separate detection and embedding (SDE) [ 3 ] directly links these three sub-tasks sequentially. However, a notable drawback arises, as these tasks cannot share any computation.…”

Section: Introductionmentioning

confidence: 99%

“…However, a notable drawback arises, as these tasks cannot share any computation. This limitation results in a disproportionately long processing time for the entire system, which prompts the exploration of joint detection and embedding (JDE) [ 3 ] architecture. The JDE methods integrate detection and ReID feature extraction into a unified model, mitigating the need for re-computation.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

PSMOT: Online Occlusion-Aware Multi-Object Tracking Exploiting Position Sensitivity

Zhao,

Zhang,

Zhang

2024

Sensors

View full text Add to dashboard Cite

Models based on joint detection and re-identification (ReID), which significantly increase the efficiency of online multi-object tracking (MOT) systems, are an evolution from separate detection and ReID models in the tracking-by-detection (TBD) paradigm. It is observed that these joint models are typically one-stage, while the two-stage models become obsolete because of their slow speed and low efficiency. However, the two-stage models have naive advantages over the one-stage anchor-based and anchor-free models in handling feature misalignment and occlusion, which suggests that the two-stage models, via meticulous design, could be on par with the state-of-the-art one-stage models. Following this intuition, we propose a robust and efficient two-stage joint model based on R–FCN, whose backbone and neck are fully convolutional, and the RoI-wise process only involves simple calculations. In the first stage, an adaptive sparse anchoring scheme is utilized to produce adequate, high-quality proposals to improve efficiency. To boost both detection and ReID, two key elements—feature aggregation and feature disentanglement—are taken into account. To improve robustness against occlusion, the position-sensitivity is exploited, first to estimate occlusion and then to direct the post-process for anti-occlusion. Finally, we link the model to a hierarchical association algorithm to form a complete MOT system called PSMOT. Compared to other cutting-edge systems, PSMOT achieves competitive performance while maintaining time efficiency.

show abstract

Learning task-specific discriminative representations for multiple object tracking

Cited by 10 publications

References 48 publications

ResNet50-deep affinity network for object detection and tracking in videos

ResNet50-deep affinity network for object detection and tracking in videos

A new detection algorithm for alien intrusion on highway

PSMOT: Online Occlusion-Aware Multi-Object Tracking Exploiting Position Sensitivity

Contact Info

Product

Resources

About