Data was collected from April until June during the darkest hours with a custom Arduino MEGA project. Street light intensity was measured using three digital luminosity sensors TLS2561 mounted on a rigid box, collecting light from the top, left, and right sides of the box (defined in relation to the vehicle’s motion). A GPS GS407A logged the location and the temperature sensor DHT22 could be used to calibrate the other sensors. Temperature, position, and luminosity from three direction were collected every two seconds on a micro-SD card.
A Python script was used to process the 33,581 points collected, with the goal to remove invalid or corrupted data, filtering adjacent points closer than 3 m, and computing the direction of the car for each point.
Street light intensity and accessibility values were normalised to allow a comparison. As the topological features of street light intensity are defined point-by-point while for accessibility are defined along a line, we used a kernel density analysis (with a 50 m radius) to estimate their density on a grid surface made by 10×10 metre cells. Once these values were placed on the grid, we computed the absolute distance between the corresponding normalized light intensity and accessibility. We used these data to highlight correlations between street light intensity and accessibility.