We work on various systems projects. Below please find a list of demo videos we have produced over years. More details about our research projects can be found at the project pages from the pull-down menu above.
Multi-level Feature Driven Storage Management of Surveillance Videos
Surveillance videos in smart environments have become commodity nowadays, which enable many novel applications, including various video analytics that turn videos into semantic results.
Different from on-demand video streaming servers, a surveillance video storage server has limited space and must retain as much information as possible, while reserving sufficient space for incoming videos. In this project, we design, implement, optimize, and evaluate a multi-level feature-driven storage server for diverse-scale smart environments, which can be buildings, campuses, communities, and cities. We focus on the design and implementation of the storage server and solve two key research problems in it, namely: (i) efficiently determining the information amount of incoming videos and (ii) intelligently deciding the qualities of videos to be kept. The goal is to retain video clips with the highest information amounts and selectively downsample the stored video clips to make room for future ones. This is not an easy task for the following reasons:
- Different video clips contain diverse information amounts, which depend on the dynamic query demands of video analytics from end-users.
- Different downsampling approaches lead to a diverse amount of information loss.
- Quantifying the information amount requires executing video analytics and downsampling video clips requires video transcoding. Both video analytics and downsampling are computationally intensive and thus need to be carefully scheduled
We use real surveillance videos from our smart campus testbed to evaluate our system. The dataset is public. However, considering the privacy issue, we blur the faces of passersby. If you want to access the dataset, please finish this application form and email it to mtsai.nthu@gmail.com
The dataset is only used for academic research. Please don’t use it for any commercial purposes.
Smart Campus Streaming Platform Demo
Along with other environmental sensors, several cameras are installed on the street lamp poles around the Computer Science building. The platform enables dynamically deploying analytics to analyze the streaming data.
Three Smart City Applications Demo
We design 3 applications, such as air pollution monitor, sound identification, and object detector. These applications demonstrate the features of our fog computing platform, such as (i) event trigger action and (ii) dynamic deployment.
Bear City Demo: Distributed Analytics
The distributed analytics accelerate the time of complex and huge operations.
Object Detection Demo: Event Trigger Action
The event trigger action saves the computing and storing resources, but pay the latency.
Fog Computing Platform
This video introduces the architecture and testbed of our for computing platform.
