Research on Pedestrian Detection and DeepSort Tracking in Front of Intelligent Vehicle Based on Deep Learning

Chen, Xuewen; Jia, Yuanpeng; Tong, Xiaoqi; Li, Zirou

doi:10.3390/su14159281

Cited by 22 publications

(9 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the DeepSort network has high accuracy and speed in multi-object tracking, it is mostly used for pedestrian and vehicle tracking and counting. It can achieve good results in tracking objects with relatively large targets and obvious features, but it is seldom used for insect trajectory tracking and counting ( Chen et al., 2022 ; Zhang, 2022 ). This is because insects are small in size, relatively inconspicuous in features, their trajectories are much smaller than those of straight-line vehicles and pedestrians, and there is no obvious motion pattern.…”

Section: Methodsmentioning

confidence: 99%

A series of methods incorporating deep learning and computer vision techniques in the study of fruit fly (Diptera: Tephritidae) regurgitation

Zhou,

Zhan,

Xiong

2024

Front. Plant Sci.

View full text Add to dashboard Cite

In this study, we explored the potential of fruit fly regurgitation as a window to understand complex behaviors, such as predation and defense mechanisms, with implications for species-specific control measures that can enhance fruit quality and yield. We leverage deep learning and computer vision technologies to propose three distinct methodologies that advance the recognition, extraction, and trajectory tracking of fruit fly regurgitation. These methods show promise for broader applications in insect behavioral studies. Our evaluations indicate that the I3D model achieved a Top-1 Accuracy of 96.3% in regurgitation recognition, which is a notable improvement over the C3D and X3D models. The segmentation of the regurgitated substance via a combined U-Net and CBAM framework attains an MIOU of 90.96%, outperforming standard network models. Furthermore, we utilized threshold segmentation and OpenCV for precise quantification of the regurgitation liquid, while the integration of the Yolov5 and DeepSort algorithms provided 99.8% accuracy in fruit fly detection and tracking. The success of these methods suggests their efficacy in fruit fly regurgitation research and their potential as a comprehensive tool for interdisciplinary insect behavior analysis, leading to more efficient and non-destructive insect control strategies in agricultural settings.

show abstract

Section: Methodsmentioning

confidence: 99%

A series of methods incorporating deep learning and computer vision techniques in the study of fruit fly (Diptera: Tephritidae) regurgitation

Zhou,

Zhan,

Xiong

2024

Front. Plant Sci.

View full text Add to dashboard Cite

show abstract

“…Yun et al [45] performed image augment by cutting and pasting masking blocks on the training image， so that the negative effect of uninformative pixels can be avoided during the training process, making the training more effective. [46]- [48] incorporated different attention mechanisms in models to capture global and rich contextual information. The visible part is fully utilized for detection, thereby effectively reducing the effect of occlusions.…”

Section: Occluded Object Detectionmentioning

confidence: 99%

YOLOv8n_BT: Research on Classroom Learning Behavior Recognition Algorithm Based on Improved YOLOv8n

Liu,

Jiang,

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Classroom learning behavior recognition can provide effective technical support for teaching and learning. However, in natural classroom teaching scenarios, classroom learning behaviors are often missed or falsely detected due to character occlusion and the small object. To tackle the above issues, this study proposed an improved classroom learning behavior recognition algorithm (YOLOv8n_BT) based on YOLOv8n. On the one hand, for the occlusion problem of classroom learning behaviors, this study incorporated the BRA into the Backbone to better capture feature information; on the other hand, for the small object problem of classroom learning behaviors for back-row-students, this study expanded a Tiny Object Detection Layer (TODL) to detect small targets better. Experiments show that the BRA and the TODL can significantly improve the model performance. The YOLOv8n_BT model, which incorporated both the BRA and the TODL into the YOLOv8n(baseline) model simultaneously, has the most significant performance improvement. Compared with the YOLOv8n(baseline), the YOLOv8n_BT model improved by 3.0%, 6.7%, 5.0%, 3.6%, and 9.0% on P, R, F1, mAP50, and mAP50-90, respectively. The detection performance of YOLOv8n_BT also outperforms other state-of-the-arts.

show abstract

“…The system was evaluated on three commonly used multi-object tracking datasets and achieved good tracking results. In their study, referenced as [24], the authors enhanced the YOLO network, devised the architecture for the DeepSORT pedestrian tracking method, and incorporated the Kalman filter algorithm for precise motion state estimation of pedestrians. The validation results show that the method reduces pedestrian targets' leakage and false detection rates.…”

Section: Related Workmentioning

confidence: 99%

Dark-SORT: Multi-Person Tracking in Underground Coal Mines Using Adaptive Discrete Weighting

Wang,

Li,

2023

IEEE Access

View full text Add to dashboard Cite

Tracking-by-detection is a popular paradigm for Multi-Object Tracking (MOT), but the problems of unstable tracking and frequent ID transitions still occur due to the low illumination, point light sources, and high dust in the underground coal mine space. In this respect, this paper proposes a Dark-SORT personnel tracking algorithm for downhole environment characteristics. First, a video image enhancement method is designed to enhance the video image quality and improve the localization accuracy of the detector for the dim and unevenly distributed light environment in the well. Second, an Adaptive Discrete-weighted Attention Module (ADAM) is designed, which consists of an Enhanced Discrete Channel Attention (EDCA) module and an Adaptive Discrete Spatial Attention (ADSA) module. EDCA enables the network to capture richer information at different scales by adaptively processing different channels according to their importance and feature scales. The ADSA approach enhances the linkage between different locations within the same region, combines different pooling strategies to highlight important regions, and reduces the focus on overexposed regions. Finally, the OC-SORT tracking algorithm is introduced to solve the error accumulation problem based on the motion model and incorporate the appearance feature information to improve the stability of target tracking. We conducted a comparison test on the self-built dataset MINE-MOT, and the HOTA, MOTA, DetA, AssA, IDF1, AssRe, and FPS metrics of the Dark-SORT tracking algorithm based on the YOLOv7 target detection model were 67.4, 92.6, 80.3, 46.8, 61.7, 65.7 and 23, respectively, which was the best in terms of accuracy and stability of all the models involved in the test.INDEX TERMS Attention mechanism, computer vision, target detection, target tracking, image enhancement.

show abstract

Research on Pedestrian Detection and DeepSort Tracking in Front of Intelligent Vehicle Based on Deep Learning

Cited by 22 publications

References 17 publications

A series of methods incorporating deep learning and computer vision techniques in the study of fruit fly (Diptera: Tephritidae) regurgitation

A series of methods incorporating deep learning and computer vision techniques in the study of fruit fly (Diptera: Tephritidae) regurgitation

YOLOv8n_BT: Research on Classroom Learning Behavior Recognition Algorithm Based on Improved YOLOv8n

Dark-SORT: Multi-Person Tracking in Underground Coal Mines Using Adaptive Discrete Weighting

Contact Info

Product

Resources

About