Data Privacy Course

One example use is bio-terrorism surveillance. Many biological agents of grave concern are not immediately noticeable in people once infected. Instead, persons live with the agent and the site remains contaminated, infecting others. Many in public health have constructed "early detection systems" in which people are observed through the data they leave behind (prescriptions, physician visits, hospital admissions, school absenteeism, etc.) in order to determine if an unusual number of people are acting ill. Thereby alarming public health early. Read the Proposal to Use Publicly Available Cameras for Automated Bio-terrorism Surveillance.

Projects in this track will record images from publicly available webcams that display images of sidewalks or roadways, and report the volume of pedestrians or vehicles appearing each hour of day (or night, as appropriate). Doing so, you will build a conceptual version of a smart camera and report whether the results are sensitive enough for bio-terrorism surveillance.

CameraWatch
The CameraWatch database provided by Professor Sweeney on-line provides a sample of URLs of publicly available webcams observing public spaces. You may have used this database in Lab 5.

Image Processing Software
An image processing program written in Java is available for manipulating images. This program is provided by the National Institutes of Health and is called Biomedical Imaging In Java. This program includes many basic routines for image manipulation in an interactive, menu-driven application. A version of this program (Zipp'ed for Mac and Unix) is available here (8MB) and an automated installation executable for Windows is available here (8 MB).

Face Detection Algorithm
Ralph Gross has provided on-line access to Henry Schneiderman's face detection program. Given a daylight image of pedestrians, for example, the face detection algorithm attempts to identify each human face appearing in the image and reports a total count. Below is a sample image from a publicly available webcam.

click for enlarged view

To use this program, you FTP your captured image into a private space. When you check back within a minute, a text file containing the results will appear in the directory. You can repeat this process as needed. Ralph Gross can provide you with username/password information for this service if you are interested in working on a project in this track; or, send a message to paddataprivacylab.org.

• Activity A
  Continuing your work from Lab 5's assignment, improve your algorithm for counting cars at night and demonstrate your results on your prevously stored images. Then, report on the performance of your improved algorithm when executed all the images from different sites that were used by students in the Lab. These images are in the FTP space. Analyze the conditions in which your solution works and does not work. This will require you to count the actual number of cars appearing in the images so you can know whether you have false positives or missed results (false negatives). Be sure to list any setup or customization needed as you go from one image to another and across cameras.

  Your project presentation will describe your algorithm and provide some examples. Vistors to your project should be able to view images and see the program working on an image. Be sure to also describe shortcomings.

  
  

• Activity B
  Improve your results from Activity A so that little or no customization is needed as you go from one image to another or across cameras.

  
  

• Activity C
  Perform surveillance on cars at a webcam site for several nights (for about 5-7 nights). Your goal is to report the number of cars appearing over time. One strategy, for example, might be to capture an image every half hour from sundown to sunrise. An alternative strategy might be to only capture images from 5-7pm each night. Plot the number of cars observed over time. You should be able to know how the traffic pattern changes over time at that location during the time observed. In order for this to work well, you need to make sure the images are captured at regularly timed intervals. Compare your computed results with the actual results (yes, you have to actually count the cars in a reasonable sample of images!).

  Your project presentation will show the traffic pattern for the location at the webcam location at night. Determine whether there is enough regularlity that if some unusual happens (such as an accident, bad weather, or a terrorist attack) whether your algorithm would be able to predict the difference. For best results, select a webcam location that you believe has a regular traffic pattern during the time of your observations.

  
  

• Activity D
  Lab 5 assignment, but instead of tracking cars at night, use the face detection program to track people during the day. You will want to select 2 webcams that show pedestrians during the day. Analyze the conditions in which the detection program works and does not work. This will require you to count the actual number of faces appearing in the images so you can know whether the program had false positives or missed results (false negatives).

  Your project presentation will describe the performance of the face detection algorithm and provide some examples. Vistors to your project should be able to view images and see the program working on an image. Be sure to also describe shortcomings.

  
  

• Activity E
  Perform surveillance on pedestrians at a webcam site for several days (for about 5-7 days). Your goal is to report the number of people appearing over time. One strategy, for example, might be to capture an image every half hour during sunlight. An alternative strategy might be to only capture images from 6-10am each morning. Plot the number of people observed over time by the face detection program.. You should be able to know how the human traffic pattern changes over time at that location during the time observed. In order for this to work well, you need to make sure the images are captured at regularly timed intervals. Compare your computed results with the actual results (yes, you have to actually count faces in a reasonable sample of images!).

  Your project presentation will show the human traffic pattern for the location at the webcam location. Determine whether there is enough regularlity that if some unusual happens (such as an accident, bad weather, or a terrorist attack) whether the program would be able to predict the difference. For best results, select a webcam location that you believe has a regular traffic pattern during the time of your observations.