A major risk we could face is the lack of datasets that cover waste that cannot be disposed of in the trash. There are many datasets of recyclable materials available online, but none that collect only images of “special waste”, although there are datasets that contain some items that fall into this category, like “Electronics Object Image Dataset | Computer Parts” on Kaggle. A mitigation strategy is to rely on YOLOv7’s own detection capabilities. YOLO should be able to detect common items, such as batteries (although testing will be needed to confirm). Another potential avenue is to make our own dataset. Python packages like simple-image-download will download images directly from Google images given a search term. We can use this to search the internet for items that aren’t present in any datasets that we want our system to reject (ie. plastic bags). We will still have to annotate the images, but software exists to make drawing the bounding boxes that YOLO needs easy, so making the dataset is not as difficult as we initially thought.
In addition to using an ultrasonic sensor to see if items have been placed on the platform, we decided to also include a weight sensor. This will not only make our design more accurate in detecting whether or not objects are present, but it will also improve our accuracy when it comes to calculating what percent of the items thrown away are trash, recycling, and other. Previously, we were calculating this by counting each individual item, but this data may not accurately reflect the actual statistics of the items thrown away, as users can recycle a whole stack of papers, and that would count for the same amount as throwing away a bottle cap. With the added weight sensor, we will calculate the percentage of recycled vs. non recyclable items by weight to allow for a more accurate user experience. The only additional cost that this will add is the cost of the weight sensor, which is $6.99. With the addition of this cost, we are still well within our budget for the project.
No changes have occurred to the schedule.
This is the display screen that we are planning to attach to the recycling bin. It currently only displays the percentage of recycled vs. non-recycled items with stubbed data. The camera only captures part of what is displayed on the screen at a time due to the shutter speed.
