A Zone-Based Multiple Regression Model to Visualize GPS Locations on a Surveillance Camera Image
Surveillance cameras are integral in assisting law enforcement by collecting video information that may help officers detect people for whom they are looking. While surveillance cameras record the area covered by the camera, unlike humans, they cannot "understand" what is happening. My research uses multiple curvilinear regression models to accurately place differentially corrected GPS points with submeter accuracy onto a camera image. Optimal results were achieved after splitting the image into four zones with the focus on calibrating each area separately. This resulted in adjusted R2 values as high as 99.8 percent, indicating that high quality GPS points can form a good manual camera calibration. To ascertain whether or not a lesser quality GPS point associated with a social media application would allow location of the person sending the message, I used an iPhone 5s to do a follow up. Using the zone-based calibration equations on GPS point locations from an iPhone 5s show that the locations collected are less accurate than differentially corrected GPS locations, but there is still a decent chance of being able to locate the correct person in an image based off that person's location. That chance, however, depends on the population density inside the image. Pedestrian density tests show that about 70-80 percent of the phone locations in a low-density environment could be used to locate the correct person that sent a message while 30-60 percent of the phone locations could be used in that manner in a high-density environment.