Learning-Based Tracking-before-Detect for RF-Based Unconstrained Indoor Human Tracking
Authors: Zhi Wu, Dongheng Zhang, Zixin Shang, Yuqin Yuan, Hanqin Gong, Binquan Wang, Zhi Lu, Yadong Li, Yang Hu, Qibin Sun, Yan Chen
IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate Neural TBD, we collect an RF-based tracking dataset in unconstrained scenarios, which encompasses 4 million annotated radar frames with up to 19 individuals acting in 6 different scenarios. Neural TBD realizes a 70% improvement in performance compared to conventional TBD methods. We first preform group-wise shuffle on RF-UNIT and divide data into train, validation, and test subset, following 8:1:1 ratio. All evaluations are reported on test sets. |
| Researcher Affiliation | Academia | 1 School of Cyber Science and Technology, University of Science and Technology of China 2 Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei, China wzwyyx@mail.ustc.edu.cn, dongheng@ustc.edu.cn, {zxshang, yuanyuqin, hanqin gong}@mail.ustc.edu.cn, {wbq0556, zhilu}@ustc.edu.cn, yadongli@mail.ustc.edu.cn, {eeyhu, qibinsun, eecyan}@ustc.edu.cn |
| Pseudocode | No | The paper describes the architecture and components of Neural TBD, including mathematical equations, but it does not include a formal pseudocode block or algorithm listing. |
| Open Source Code | Yes | The code and dataset will be released. The dataset and code will be public. |
| Open Datasets | Yes | To evaluate Neural TBD, we collect an RF-based tracking dataset in unconstrained scenarios, which encompasses 4 million annotated radar frames with up to 19 individuals acting in 6 different scenarios. We present the RF-UNIT dataset, which encompasses million-level radar heatmaps of at most 19 individuals in multiple different scenarios. The dataset and code will be public. |
| Dataset Splits | Yes | We first preform group-wise shuffle on RF-UNIT and divide data into train, validation, and test subset, following 8:1:1 ratio. |
| Hardware Specification | Yes | All experiments are conducted on a single NVIDIA A100 GPU with a batch size of 16. |
| Software Dependencies | No | The paper mentions using the Adam optimizer, but it does not specify software dependencies with version numbers for libraries or frameworks used (e.g., PyTorch, TensorFlow, Python version). |
| Experiment Setup | Yes | We employ the Adam optimizer with initial learning rate of 1.0 10 2 and weight decay of 0.05. During the training process, we adopt a step-based learning rate decay strategy. All experiments are conducted on a single NVIDIA A100 GPU with a batch size of 16. |