students: anthony kang, luke neumann, erik rozolis professors: dr. edward j. delp, dr. yung-hsiang...
TRANSCRIPT
Students: Anthony Kang, Luke Neumann, Erik RozolisProfessors: Dr. Edward J. Delp, Dr. Yung-Hsiang Lu
CAM2Continuous Analysis of Many CAMeras
The ProblemCurrently there is no common computing infrastructure which can support continuous analysis of many internet connected cameras and so researchers are unable to harness this vast amount of data for analysis.
The Infrastructure1. A large number of publicly available cameras and their properties2. A set of functions common to different analysis programs and an API for developing
the analysis programs3. A resource manager (system) which optimizes the resources needed for running the
analysis programs
Possible ApplicationsUses of the system include:• Traffic analysis to improve congestion or detect accidents• Weather observation to increase the accuracy of existing weather models• Environmental trends (rising sea levels, shorter days, etc.)
System Team Progress• Database improved for easier data management• System redesigned to handle a more diverse set of
applications• Locations of cameras found so far mapped and clustered
(below)• Implemented Django for better website management
and easier modification• Camera search given more specific options for faster
loading• Query improved for faster map loading
Cloud-Based Distributed System ArchitectureAbove is a lower level view of the interactions of the distributed system from application upload to the job scheduler. Two monitors provide information about the performance of the system and the data flow during operation. A database holds everything from logins for the website to reports generated automatically by the manager.
Image Processing APIThe image processing API is designed for image processing algorithm testers to use our system easily. All they need to do is just write a pure processing algorithm and plug it into the API.
Tasks for the API1. Download images for cameras and store them into a buffer.2. Send Reports to the system.3. Run the image processing algorithm.
Future DevelopmentShort Term:• Improve camera search function• Scale up from 7 worker computers and 30,000 cameras to
hundreds of workers and hundred thousands of cameras• Build API which will interface with a website• Improve rain detection and sunrise/sunset algorithmLong Term:• Scale up to millions of cameras and thousands of workers• Upgrade to billable system• Develop 4 or 5 weather detection algorithms
Rain Detection algorithmThis algorithm analyzes videos from cameras and detect if it is raining.
Fig.6 Weight of Rain
Fig.2 CloudyFig.3 Rain
Fig4 weight of sunny Fig.5 Weight of cloudy
Fig.1 Sunny
Sunrise/Sunset DetectionThis algorithm analyzes videos from cameras and detect if it is sun rising or sun setting. From the result retrieved, the length of the day can be calculated.
Steps1. Detect Horizon2. Calculate brightness of sky using pixels above horizon3. Store value and repeat4. Sudden increase in brightness in morning may indicate sunrise.5. Sudden decrease in brightness in evening may indicate sunset.
Time: 20:42 RGB Avg: 52Time: 20:39 RGB Avg: 71Time: 20:31 RGB Avg: 111
sunset
Rain Detection results
Automatic Testing and Integration• Use Jenkins and Github to test and integrate new code• Protects the main site from breaking due to new code