AIWaysion
AutoML for traffic video analysis system using Query-based learning
Network cameras are commonly used in both consumer and industrial settings, and they play a crucial role in enhancing traffic safety through advanced tracking-based video monitoring. In this project, our goal is to develop computer vision AI models for traffic video analysis, specifically for tasks like vehicle counting and anomaly detection. However, these AI models need to stay up-to-date as new car models and unfamiliar vehicle types emerge. This student team will work to explore technical solutions that allow AIWaysion to continuously update existing AI models through human interaction using query-based learning. AIWaysion aims to establish a dynamic connection between humans and machines, utilizing object detection data to create object trajectories for analyzing video footage. This approach can help AIWaysion customers easily comprehend traffic events and identify those that require their attention. This student team will work to conduct thorough research on cutting-edge solutions for multi-object tracking and implement state-of-the-art methods to train machine learning models. These models will be used to develop practical applications based on video recordings captured by AIWaysion devices, such as vehicle counting and anomaly detection. This student team will work to achieve the following project goals: 1. Research: This student team will work to conduct an in-depth exploration of cutting-edge algorithms and models in the field of multiple-object tracking and related subjects. 2. Software: This student team will work to identify and experiment with open-source codebases for training multiple-object tracking models. This student team will also work to develop a comprehensive pipeline for data processing, model development, and evaluation. 3. Data: This student team will work to acquire existing datasets for multiple-object tracking and pre-trained models. This student team will also work to collect dedicated data using AIWaysion devices and initiate the process of annotating the collected data to build a foundational dataset. The expected achievements this student team will work to accomplish include: 1. This student team will work to develop a comprehensive deep learning model capable of generating object trajectories from provided video inputs. 2. This student team will work to create a codebase for training and assessing the performance of this model. 3. This student team will work to establish a dataset containing tracking annotations for data captured by AIWaysion devices. 4. This student team will work to compile all project-related reports and presentations as documentation. 5. This student team will work to implement a backend system to host the models and provide support for the aforementioned functionalities. 6. This student team will work to create a frontend system to engage with users and visually present the generated results.
Faculty Adviser
Wei Cheng,
Adjunct Professor,
Electrical & Computer Engineering
Students
Chenwei Huang
Haoxiong Zhang
Mathew Garcia-Medina
Shaoyu Chen
Xinpeng Shan
You Guo