TSLNet: a hierarchical multi-head attention-enabled two-stream LSTM network for accurate pedestrian tracking and behavior recognition
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.TSLNet, a novel Hierarchical Multi-Head Attention-Enabled Two-Stream LSTM Network, enhances pedestrian tracking and behavior recognition. This advanced model excels in complex environments, improving surveillance and smart transportation systems.
Area Of Science
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background
- Accurate pedestrian tracking and behavior recognition are crucial for intelligent surveillance, smart transportation, and human-computer interaction.
- Real-world video data presents challenges like environmental variability, high-density crowds, and diverse pedestrian movements.
Purpose Of The Study
- To introduce TSLNet, a Hierarchical Multi-Head Attention-Enabled Two-Stream LSTM Network.
- To improve pedestrian tracking and behavior recognition in complex and dynamic environments.
Main Methods
- TSLNet integrates a Two-Stream Convolutional Neural Network (Two-Stream CNN) with Long Short-Term Memory (LSTM) networks for spatial-temporal feature extraction.
- A Multi-Head Attention mechanism focuses on relevant features, while Hierarchical Classifiers within a Multi-Task Learning framework enable simultaneous basic and complex behavior recognition.
Main Results
- TSLNet significantly outperforms existing baseline models on multiple datasets.
- Achieved higher Accuracy, Precision, Recall, F1-Score, and mAP for behavior recognition.
- Demonstrated superior MOTA and IDF1 for pedestrian tracking.
Conclusions
- TSLNet effectively enhances both pedestrian tracking and behavior recognition performance.
- The proposed network is highly effective in handling complex real-world video data.
- TSLNet shows significant potential for applications in intelligent surveillance and smart transportation.
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
