Shuaiqing SONG

1. PRP/012/22CI | Future Classroom with AIoT System for SEN Students

- Developed traditional computer vision behaviour analysis module for smart classroom perception

- Built pipeline with YOLO for multi-target region detection and OpenPose for human keypoint extraction

- Analysed temporal keypoint sequences and human interactions to recognise SEN students’ learning behaviours

2. 317 | Detection of Human Interaction for Visually Impaired People

- Took charge of the vision perception branch of the quadruped robot system

- Deployed off-the-shelf pre-trained vision foundation models to recognise pedestrians and human interactions

- Organised inference outputs into structured morning & afternoon command maps

- Supplied vision results for cross-modal validation with LiDAR to generate robot motion commands

3. 386 | Intelligent Specific Learning Difficulties Screening Approach for SEN Students

- Led end-to-end construction of SEN student behaviour video dataset

- Used YOLO and OpenPose to automatically filter, clip and purify valid video samples

- Participated in the full lifecycle of customised VLM reconstruction and fine-tuning

- Adopted MMAction2 action recognition framework to cross-validate VLM outputs and reduce false predictions