Fair and efficient spatial siting of public surveillance cameras

Public surveillance cameras are expensive, and residents living near them bear the cost of being watched. Siting cameras purely where crime is most concentrated can place that burden disproportionately on minority neighborhoods. I formulate camera siting as a maximal covering integer program: street locations are candidate camera sites, each camera captures outdoor crimes within 1,000 feet, and no crime is counted twice. Census block demographics are allocated to street segments with street-weighted dasymetric mapping, so the racial composition of residents living near any set of cameras is a linear function of the siting decision. A single linear constraint then caps the minority share of residents near the selected cameras. Using four years of outdoor, victim-reported crime in Dallas, Texas, I trace how many crimes 100 to 1,000 cameras can capture, who lives near the most efficient placements, and how much capture is lost when the minority share of residents near cameras is capped at progressively lower levels. Small programs cluster downtown, near residents whiter than the city overall; larger programs place a disproportionate share of Black residents near cameras. Capping the minority share at 70%, just below the city-wide 72%, costs at most about 1% of captured crime at any program size, and a 55% cap costs at most about 6%, with similar costs in a held-out year. Code, data, and paper source are at https://github.com/apwheele/FairCameraSiting.