Project Overview
This innovative project seeks to streamline surveillance by efficiently monitoring and tracing the movement of various objects using a single camera. By leveraging advanced computer vision and machine learning technologies, the system aims to provide comprehensive tracking capabilities, enhancing security protocols within confined spaces.
The project benchmarked various deep learning networks and different optimization techniques, including DeepSORT, FairMOT, Tree Parzen Estimator and feature matching algorithms. The system is a breakthrough step to achieve a more ambitious system, a Muti-Camera Multi-Object Tracking system.
The system is a breakthrough step to achieve a more ambitious system, a Muti-Camera Multi-Object Tracking system.
Responsibilities
Researched, developed and ensembled several detection models for comparative analysis (mostly YOLOv5 and CenterNet).
Research, optimized and benchmarked different tracking models (DeepSORT, FairMOT), resulted in approximately 85% in term of MOTA metric on collected dataset.
Designed partial automation pipeline with detection model to visualize and extract most valuable data partitions, resulted in much smaller and more meaningful dataset.
Leaded data team and annotation pipeline, including data collection, extraction, annotation and quality verfication.
Prepared weekly reports and represented project progress to stakeholders.