Alan Shangguan

Extending our DeepStream edge pipeline with YOLO pose estimation — keypoint tracking, metadata export, and validation on production camera feeds — to detect unsafe worker behaviour in real time.

1. Develop and deploy computer vision solutions for train station monitoring and safety enhancement, including the City Rail Link (CRL) and other major transport projects. 2. Build and optimize large-scale CCTV analytics pipelines using Triton Inference Server and OpenShift, ensuring scalable and real-time inference. 3. Integrate YOLOv8 and OCR/number plate recognition models for enhanced analytics in transport applications. 4. Manage SQL databases for data storage and reporting, supporting operational decision-making. Implement Azure CI/CD pipelines to streamline model deployment and system updates within Agile development frameworks.

1. Researched and engineered ML/CV algorithms for autonomous vehicles in vineyards, enhancing obstacle detection using camera and LiDAR data, thereby improving operational safety and efficiency. 2. Led the data collection and annotation process using CVAT, ensuring high-quality datasets for training robust semantic segmentation models with Python/PyTorch. 3. Work in agile environments, Successfully deployed semantic segmentation TensorRT models on edge devices (e.g., Jetson Xavier) for real-time processing, utilizing C++ with ROS and Docker for scalable and reliable applications in field operations..

1. Researched and engineered the development of AI-driven smoke detection solutions for early fire/smoke detection in wild and urban settings, significantly contributing to global deployment in key regions, utilizing Python and PyTorch for algorithm development. 2. Fostered collaboration within R&D teams to create versatile detection systems for diverse platforms (PC, mobile, cloud, edge), meeting client needs across governmental and corporate sectors. This included integrating solutions with Google Cloud for scalable cloud computing and deploying on edge devices such as the Jetson Nano for on-site processing. 3. Directed comprehensive data handling initiatives, from collection to labeling, employing tools like CVAT, labelImg, and labelme, to train and evaluate models (YoloV5, YoloV3, Semantic Segmentation) effectively with Python and PyTorch, ensuring robust model performance.

1. Applied machine learning techniques to segment and classify breast cancer images from H&E stained slides, enhancing diagnostic accuracy. 2. Developed a pathology image processing application, bridging the gap between computer vision technology and medical needs. 3. Collaborated closely with the Department of Medicine, showcasing the ability to work effectively in interdisciplinary teams.

Developed a video player website and a large-size video uploader, integrating frontend and backend technologies (HTML, JavaScript, AJAX/jQuery, CSS, ASP.Net Webforms, SQL Server, C#), demonstrating full-stack development proficiency. 3. Using Html, JavaScript and ajax/jQuery, CSS, ASP.Net Webforms, SQL Server, and C#

Thesis: Video-based Motion Capture using AI, Single-person 3D pose estimation in RGB image