Our main goal of this project was to automate the process of detecting tobacco ads in an urban environment.  Automation reduces the amount of resources it takes to collect data on where signs are located. Improving on the accuracy of our model would allow us to also perform a longitudinal study as new images are captured for the same locations in order to analyze how the marketing practices of tobacco companies changes over time in the city, particularly in response to new regulations. As we continue our work, there are several methods we plan to employ to improve the precision of our detection algorithm. Collecting high resolution images, using the Google Street View premium service for example, would support human recognition of additional ad during the labeling process. Furthermore, a larger original training dataset would support our model in identifying unique characteristics of tobacco ads for use in classifying ads from yet-to-be-seen images of storefronts. We collected over 40,000 images but only had time to visually inspect about half of them. Finally, higher confidence in our detection of cigarette ads in the city using automated methods would be encouraging for detecting other urban features as well. Similar studies could be undertaken to determine the impacts of alcohol advertising or fast food displays, and provide another demonstration of how emerging technologies might be used beyond commercial applications to help improve our communities.
Because our model deserves improvement, the socioeconomic analysis map stands as a  proof of concept. Upon strengthening the confidence of our detections, we plan to re-measure the exposure of children to tobacco advertising and expand our analysis to include scrutinization as to how companies advertise to different socioeconomic groups. As a result, we think it's quite useful as an exploratory analysis and begs the question about how the tobacco industry might be targeting various parts of NYC differently (prices, brands, products, marketing campaigns). Furthermore, it will be insightful in future work to overlay this data with school locations and likely trajectories to examine exposure risk specifically for school children on their way to, from, and around school. Big Tobacco clearly has a sense of where their target demographics reside and spend their time, but the breakdown probably isn't as straightforward as a spatial distribution of population density or median income.