The Community Science project I participated in was “Industrial Smoke Hunting” on scistarter. This project was presented by a cohort of the Robotics Institute at the Carnegie Mellon University. “Industrial Smoke Hunting” aims to distinguish smoke fumes from video clips of industrial sites. While Artificially Intelligent systems aim to do this type of work independently, human presence is required to make those systems possible. As such, my participation involved manually identifying fumes in video clips. I felt that this project relates to a more realistic outlook on the dystopian threats of AI in the present day, where such technologies are still fairly primitive but can be helpful in research. Understanding when and where smoke fumes are detected is the prerequisite for other research to take place. 

Heatmaps used to identify smoke plumes.

What caught my attention to this project was the blatant complexity of distinguishing between natural elements – like clouds – from smoke. A plethora of contemporary computer vision systems have nuanced features to filter out natural features that may impact the detection of an object but are not subject to the various luminance and weather conditions that may alter the appearance of a cloud of smoke on an imaging sensor.  I ultimately realized that proper camera placement was important and it helped that the videos served had cameras at vantage points that had contrasting natural features.

Further, this project reminded me of the science poster my group and I created, where we manually identified geographical coordinates in google earth and used human discretion to classify the location.  Our three classifications were based on somewhat subjective factors. While classifying our data set, it became clear that some of the repetition and subjectivity could be eliminated by technology. In order for that to be possible, we would need to first define our classifications based on a test set of imagery, which would likely require a far more extensive process than manual identification, as we would need to create comprehensive models for each classification (which in turn would require more manual identification of geographical features),

All in all, I found this to be an interesting way to partake in community science, and those interested in joining as well can refer to the project’s web page at : https://scistarter.org/industrial-smoke-hunting