-
Connecting researchers with data for conservation
Inspiration / Challenge
With the COVID-19 pandemic, more people than ever are staying inside their homes. While staying at home is important for reducing the spread of coronavirus, it doesn't mean we need to be locked up in our home every minute of the day. We still need to be exercising and going outside to keep our sanity.
Current fitness tech encourages people to stay indoors (Peloton, online classes), or simply tracks data and provides analytics. What’s more, there are plenty of people who hate exercise but want to find a way to get it done.
Our fitness tech promotes the most basic form of exercise in a way where you don’t even know you’re exercising for heart health — walking. To give people more excuses to enjoy the outdoors, we engage people with a favorite pastime — taking photos. In doing so, we designed it so that the pictures taken ultimately benefit researchers as well as the environment.
What it does
What our app ultimately does is crowdsource research to make data available for researchers and others interested in combating climate change. Not only can users take photos of fauna, flora, or other items for any number of projects listed by researchers and then upload images to our database that come along with geolocation and other metadata, but they can also learn about the projects they’re contributing and why what they’re doing is important. In turn, researchers can make use of data in several ways.
First, when users upload images of birds in their local area, for example, they help capture trends in biodiversity from what species are present and when. Researchers can then observe the decline or even rise of bird species in an area or observe migratory patterns from which to draw more questions and research further. There are so many important bird areas where people go birdwatching and take close up shots of nature, so why not contribute to research while you’re at it?
Second, when users upload images of a tree in their area that they may have adopted — similar to how people adopted fire hydrants — they help nonprofit organizations like Speak for the Trees and cities create a tree inventory that can be used for planning tree canopy. Urban forestry planning can influence people’s everyday lives by reducing stormwater runoff, decreasing wildfire risk and severity, reducing urban heat islands, decreasing utility costs, increasing economic growth, and providing clean drinking water. Plus, trees can sequester atmospheric carbon dioxide (CO2) and serve as long-term carbon sinks.
Third, users can even report and upload images of road closures, impromptu construction sites, or accidents to help redirect traffic and avoid having idling automobiles generating pollution and wasting gas.
The possibilities for use cases are endless—
Why crowdsource research in the first place?
Tasks that may have previously been conducted by a small team of researchers can now be parallelized and processed by millions of volunteers over the Web, making questions that seemed previously impossible now tractable.
Crowdsourcing research leverages the ease with which anyone can upload content to the internet, helps eliminate biases in data collection that can happen location-wise, time-wise, etc., and makes use of people’s knowledge of and access to their local communities. In allowing researchers to gather data, they put funding towards other important items or make up for what grants have not been able to fund.
How we built it
We used HTML and JavaScript in the front end, Python Flask in the back end, DataStax for the database, and GCP Vision AI for training a model to identify 200 species of birds.
Challenges we ran into
We initially tried implementing Django for the back end and ended up going with Flask.
Learning new technologies like connecting our project to the DataStax database was a learning curve. We also learned that files can’t be moved across regions in GCP. Instead we had to use a regional bucket in the same location and with same storage class as AutoML.
Accomplishments that we're proud of
Our final product looks clean!
What we learned
We executed some nifty scripts to feed in Kaggle data to train the model on GCP AI, as we can’t imagine writing 30,000+ rows of file paths and category labels.
What's next for Hivemind
For ease of use, we could provide higher-level categories of potential actions when people first sign in such as “Take pictures of nature”, “Adopt a tree for the day”, “Make someone’s day” and allow people to bookmark their favorite activities. After clicking on an activity, either the back end knows how to sort what photos are relevant to what projects, or it breaks down at a high level what type of activity people want to pursue (e.g. for make someone’s day, users can report road closures or call out the weather forecast).
Log in or sign up for Devpost to join the conversation.