Video Analytics for Daily Living (VADL) labSchool of Electrical and Computer Engineering
Video analytics for agricultural machinery
The target of this project is to design and develop systems for farming machinery automation. Typical farming machines include tractors, combine harvesters, and choppers. Cameras installed inside the cockpit of these vehicles capture the surroundings. Using video analytics, the system interprets and logs information about nearby activities. This work was supported by the Open AG Technology and Systems (OATS) center.
Video analytics for dairy cow health assessment
In this project, we monitor a cow's health and well-being using surveillance camera videos. Multiple cameras are installed in a local dairy farm to observe cows walking as they exit the milking parlor. By analyzing the cow walking videos, we extract high-level information about each cow, such as identity, weight, and lameness. This work is supported by the Open AG Technology and Systems (OATS) center.
Video analytics for turkey welfare
Turkey production is very important in the United States. Production can be impacted by turkey welfare. However, changes in turkey welfare might not be completely visible to humans. Therefore, we work on building a robust system that is capable of tracking turkeys and identifying different types of aggregated behavior. Building such a system requires the usage of video analytics, computer vision, deep learning, and image processing techniques. By successfully analyzing and detecting changes in turkey welfare, farmers or researchers can take appropriate actions before problems further escalate.
Image and video quality assessment for video analytics
Video quality and video content both affect the performance of emerging video-analytic tools, or tasks. Factors such as compression, lighting, and an individual's distinctiveness all have significant impact on task performance. In this project, we explore task-based video quality assessment.
Video analytics for assessing hand-hygiene
This project focuses on assessing the quality of hand-hygiene activities for the food safety industry. Effective hand-hygiene can reduce food contamination by human pathogens at during all stages of food handling. We use image processing, computer vision, and deep learning methods to build a system for hand-hygiene activity recognition. Analyzing the input hand-hygiene video, the system reports all the hand-hygiene actions performed by the user.
Joint Video Compression and Analytics
Practical video analytics systems that perform computer vision tasks are widely used in critical real-world scenarios such as autonomous driving and public safety.
These systems are usually deployed in bandwidth-constrained environments; hence, popular video compression algorithms like HEVC compress the inputs which are then passed onto modules that perform tasks like object detection, segmentation, and recognition sequentially.
Video compression reduces transmission bandwidth but also degrades system performance. Here, we focus on understanding the effects of compression on the overall system performance using meaningful and interpretable evaluation procedures. We utilize this information to develop strategies that potentially reduce the impact of compression on system performance.