P-IoU: Accurate Motion Prediction Based Data Association for Multi-object Tracking

Wu, Xinya; Xu, Jinhua

doi:10.1007/978-981-99-8073-4_37

Cited by 2 publications

(4 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Predictive Tracking. Most tracking by prediction paradigms are different from ReID-based methods, such as [6,7,17,18], which aim to discard ReID association which is computational and not very stable when obejct occluded and tracking uniform targets [23]. SiamMOT [7] introduces two kinds of motion models to estimate the motion of bounding boxes for targets in the previous frame.…”

Section: Relative Workmentioning

confidence: 99%

“…Given that deep detection networks could already adapt well to complex scenes, popular tracking algorithms primarily focus on enhancing the tracking aspect. This involves optimizing target association methods [15,16], IOU (Intersection over union) matching [1,2,6], leveraging prior cues from motion models [5][6][7]17], and refining target texture features [2,4,18]. These association methods are typically combined, with texture features often serving as the primary feature for target matching, being directly employed in matching targets and trajectories.…”

Section: Introductionmentioning

confidence: 99%

“…The motion model plays a crucial role and is typically employed in two ways. Firstly, they involve directly predicting the target's bounding box position in the next frame and then applying spatial matching with the detected bounding box [6,7,17]. This motion model bypasses the texture feature matching process, saving space and reducing complexity.…”

Section: Introductionmentioning

confidence: 99%

“…falls below a specified threshold [6,17]. Beyond its role in assisting motion model and texture feature matching, IOU matching contributes to confirming unverified tracks by matching trajectories and detections that remained unmatched in the preceding matching step [1,6].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Application of ROI Feature Decomposition and Local Matching in Multi-Object Tracking

Yang,

Ban,

Liu

2024

Preprint

View full text Add to dashboard Cite

For multi-object tracking (MOT), matching newly detected target features and trajectory features often involves template matching. Typically, feature embedding for multi-object tracking relies on either the mean feature within the Region of Interest (ROI) or the vector at the target's center. However, both of these feature embeddings are vulnerable to interference in scenarios involving occlusion. Additionally, one-to-many (O-T-M) style matching strategies are prone to causing the loss of trajectories and detections, as one target only represented by a single indicator. To overcome those limitations, we propose utilizing Principal Component Analysis (PCA) on the target ROI to decompose target features, expressing the identical feature of targets by collectively representing the principal component features of the ROI region, and utilizing these principal component features to form local features, the identical features are then trained through a classification task. Furthermore, in the template matching process, we introduce a local features matching algorithm for targets and trajectories. This algorithm adopts a many-to-many (M-T-M) matching style, aiming to associate every detailed part of the ROI feature. Experimental results demonstrate that the proposed local matching algorithm effectively handles complex and dynamic scenes. The approach achieves leading performance on the motchallenge benchmark, validating its effectiveness.

show abstract