LSTM guided ensemble correlation filter tracking with appearance model pool

Jain, Monika; Subramanyam, A V; Denman, Simon; Sridharan, Sridha; Fookes, Clinton

doi:10.1016/j.cviu.2020.102935

Cited by 6 publications

(5 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Besides, Some methods adopt motion models, such as the Kalman filter (Wojke et al, 2017;Zhang et al, 2021), optical flow (Tang et al, 2017), and motion prediction networks (Zhou et al, 2020;Sadeghian et al, 2017;Wang et al, 2022), that incorporate temporal features to make dynamic position predictions to compensate for noisy detections. Some methods establish Recurrent Neural Networks (Milan et al, 2017;Sadeghian et al, 2017;Jain et al, 2020) to model complex motion patterns. Moreover, data association is also formulated as a graph optimization problem in some methods and solved globally with network flow (Schulter et al, 2017) and Multiple Hypothesis Tracking (Kim et al, 2015) frameworks.…”

Section: Tracking-by-detectionmentioning

confidence: 99%

Multi-object tracking with robust object regression and association

Li¹,

Ji²,

Chen³

et al. 2023

Computer Vision and Image Understanding

View full text Add to dashboard Cite

Section: Tracking-by-detectionmentioning

confidence: 99%

Multi-object tracking with robust object regression and association

Li¹,

Ji²,

Chen³

et al. 2023

Computer Vision and Image Understanding

View full text Add to dashboard Cite

“…For the data with video image sequence, RNN and LSTM are good choices, so LSTM is selected as the data classifier in this paper. Long short-term memory (LSTM) model is usually used for classification [24]. LSTM is an improvement in recurrent neural network (RNN).…”

Section: Introductionmentioning

confidence: 99%

Research on Safe Driving Evaluation Method Based on Machine Vision and Long Short-Term Memory Network

Shi

Tang²

2021

Journal of Electrical and Computer Engineering

View full text Add to dashboard Cite

The rapid development of transportation industry has brought some potential safety hazards. Aiming at the problem of driving safety, the application of artificial intelligence technology in safe driving behavior recognition can effectively reduce the accident rate and economic losses. Based on the presence of interference signals such as spatiotemporal background mixed signals in the driving monitoring video sequence, the recognition accuracy of small targets such as human eyes is low. In this paper, an improved dual-stream convolutional network is proposed to recognize the safe driving behavior. Based on convolutional neural networks (CNNs), attention mechanism (AM) is integrated into a long short-term memory (LSTM) neural network structure, and the hybrid dual-stream AM-LSTM convolutional network channel is designed. The spatial stream channel uses the CNN method to extract the spatial characteristic value of video image and uses pyramid pooling instead of traditional pooling, normalizing the scale transformation. The time stream channel uses a single-shot multibox detector (SSD) algorithm to calculate the adjacent two frames of video sequence for the detection of small objects such as face and eyes. Then, AM-LSTM is used to fuse and classify dual-stream information. The self-built driving behavior video image set is built. ROC, accuracy rate, and loss function experiments are carried out in the FDDB database, VOT100 data set, and self-built video image set, respectively. Compared with CNN, SSD, IDT, and dual-stream recognition methods, the accuracy rate of this method can be improved by at least 1.4%, and the average absolute error in four video sequences can be improved by more than 2%. On the contrary, in the self-built image set, the recognition rate of doze reaches 68.3%, which is higher than other methods. The experimental results show that this method has good recognition accuracy and practical application value.

show abstract

“…Discriminant methods that address tracking by discriminating the foreground from the background of an image are closely related to target detection (Li and Zheng, 2020 ), image segmentation (Bhandari et al, 2020 ; Guan et al, 2021 ), and other technologies. The boosting method (Yang et al, 2016 ), support vector machine (Feng et al, 2018 ), and deep neural network (Tong et al, 2019 ; Jain et al, 2020 ) are all good approaches for discriminant target tracking. Generative tracking methods rely on certain tracking strategies to determine the optimal solution from many candidate targets.…”

Section: Introductionmentioning

confidence: 99%

Introducing Depth Information Into Generative Target Tracking

et al. 2021

View full text Add to dashboard Cite

Common visual features used in target tracking, including colour and grayscale, are prone to failure in a confusingly similar-looking background. As the technology of three-dimensional visual information acquisition has gradually gained ground in recent years, the conditions for the wide use of depth information in target tracking has been made available. This study focuses on discussing the possible ways to introduce depth information into the generative target tracking methods based on a kernel density estimation as well as the performance of different methods of introduction, thereby providing a reference for the use of depth information in actual target tracking systems. First, an analysis of the mean-shift technical framework, a typical algorithm used for generative target tracking, is described, and four methods of introducing the depth information are proposed, i.e., the thresholding of the data source, thresholding of the density distribution of the dataset applied, weighting of the data source, and weighting of the density distribution of the dataset. Details of an experimental study conducted to evaluate the validity, characteristics, and advantages of each method are then described. The experimental results showed that the four methods can improve the validity of the basic method to a certain extent and meet the requirements of real-time target tracking in a confusingly similar background. The method of weighting the density distribution of the dataset, into which depth information is introduced, is the prime choice in engineering practise because it delivers an excellent comprehensive performance and the highest level of accuracy, whereas methods such as the thresholding of both the data sources and the density distribution of the dataset are less time-consuming. The performance in comparison with that of a state-of-the-art tracker further verifies the practicality of the proposed approach. Finally, the research results also provide a reference for improvements in other target tracking methods in which depth information can be introduced.

show abstract

LSTM guided ensemble correlation filter tracking with appearance model pool

Cited by 6 publications

References 50 publications

Multi-object tracking with robust object regression and association

Multi-object tracking with robust object regression and association

Research on Safe Driving Evaluation Method Based on Machine Vision and Long Short-Term Memory Network

Introducing Depth Information Into Generative Target Tracking

Contact Info

Product

Resources

About