Person Tracking for Parking Space VacancyPredictionMooneer SalemDepartment of Computer Science and EngineeringUniversity of California, San DiegoLa Jolla, CA [email protected] WatsonDepartment of Computer Science and EngineeringUniversity of California, San DiegoLa Jolla, CA [email protected] has a serious problem with undergraduate student parking. Most studentsmust resort to parking in far corners of the campus and taking shuttle buses to theirclasses. When arriving at campus, people generally search for vacant parkingspaces from the closest spaces to the farthest, unnecessarily wasting time. Wepropose a solution whereby students can determine whether a parking space isabout to become vacant, reducing the time spent searching. A camera is pointed ata group of parking spaces in a given lot. The faces of drivers and their passengerswill be detected through existing software, and their movements will be tracked.If the movements correspond with entry into a vehicle, then the parking space isabout to become vacant. This information will be displayed both on a publically-accessable Web site (for use by mobile phones) and possibly on a sign readablefrom a distance.1 QualificationsMooneer: I took CSE 151 (Artificial Intelligence) during Spring Quarter of 2006 and Math 183(Probability and Statistics) in Winter 2005. As for other math classes, I’m currently taking Math102 (Applied Linear Algebra), Math 170A (Numerical Linear Algebra) and have taken Math 20Fin Spring 2004. I’ve also taken and completed several major project courses, including CSE 131A(Compilers I) and CSE 141L. As for programming experience, I am experienced in Java, C, C++,and a little bit of C# on both Windows and Unix platforms.Daniel: I am a UCSD computer science student with a strong background in mathematics. Duringhighschool I skipped several grades of math and took ”Math 1D” at Foothill College during the sum-mer between my sophomore and junior year. This class covered vector fields and parameterization.In winter of 2005 I took Math 20F at UCSD, covering matrix manipuations and covering use of Mat-lab. Additionaly, I took CSE 151 in the spring of 2006 with Gary Cottrell. Since June 2006 I havebeen working at Avaak with small cameras. Through Avaak I met with Associate Professor Clark C.Guest of UCSD and started working on a project in October 2006. This project takes pictures froma camera pointed at a passive chemical sensor and extracts information about what chemicals are inthe air. Working together we will gather a database of pictures and create a formula for determiningthe content of the air. I have also been working with Preuss High School’s robotics team since 2004mentoring the robotics team. I developed the vision system that allowed the robot last year to tracka green light for aiming and firing nerf balls.2 MilestonesChristmas Break (December 10-January 7) Obtain iSight from a retailer and set up the cameraon hardware. Also, use iSight to get MPlab demo programs working during this time andfigure out how to extract pictures from the camera programmatically using Windows orLinux driver. Determine best way to put up a sign to alert drivers of spaces that are openingup.Weeks 1-3 (January 8-26) Begin program that will perform the detection. Develop algorithm todetect movement of any object in the camera’s field of view and write code to updatesign(s) as needed.Weeks 3-5 (January 22-February 9) Programmatically detect people’s faces using MPlab’s li-brary. Using location derived from MPlab in each image captured, detect the directionof movement of individuals to and from their cars.Weeks 5-6 (February 5-16) Using human tracking features, detect spaces that are about to becomevacant. Upload this information to the Internet in a human-readable form.Weeks 6-7 (February 12-23) Test the system using a variety of parameters: different views, differ-ent times of day, and different people entering and leaving cars.Weeks 7-10 (February 19-March 16) Finalize research paper and present to other CSE 190 stu-dents at the end of the quarter.3 Division of LaborThere’s currently no definite division of labor decided as of yet. Both of us will work on the individ-ual components of the project together. However, Daniel will be specializing more in the computervision aspect of the project, while Mooneer will specialize in the software development side of theproject.4 Questions To Be AnsweredThe following subsections list the questions we hope to answer during the project, along with somethoughts on each of them.4.1 How will passengers and drivers be differentiated?Drivers will always enter the vehicle from the front left door (assuming the system is used in theUnited States). The tracking software will detect the direction individual people are moving and willcalculate which door the person entered based on that information.4.2 How does time of day affect the accuracy?The time of day will affect the contrast. During the night, lights will be much more visible comparedto everything else. We can simulate the darker conditions by putting a filter on the lens of the camera.4.3 What angle for the camera is optimal for determining spaces that are about to open up?We can choose from a variety of positions for the camera. Due to limitations in the MPLab software,however, the camera must face the people approaching or leaving the vehicles.4.4 Will it help to keep a history and use that to update guesses?We will attempt to keep a history of each spot and track people across time to determine their locationand direction. This extra information will be helpful in determining in a given picture whether a spotis about to be filled.4.5 How will the direction people are facing affect accuracy?The MPLab software works best when people’s faces are visible (that is, no hats or veils coveringfaces, or even walking in a direction opposite the camera). As a fallback, the system could possiblydetect people based on any kind of movement, not just the movement of confirmed faces.4.6 How can detecting any movement filter out only people?We plan on using the face detection for the majority of the decisions the system has to make. Ifno faces are detected with the MPLab software, the system should fallback to general movementdetection.4.7 How will performance be evaluated?Performance will be evaulated on the basis of whether the system correctly determines whether aparking space is about to be freed. Training