• More Testing

Since the current experiment condition is limited, we plan to do more testing on different conditions such as gas leak, heavy air pollution and so on.

​

• Maybe change to another faster database resource

We found that DynamoDB may not be a good database resource because it is too slow to store or load our data within few seconds. For further work, we may use another faster database resource to store our data.

​

• Switch to one board implementation only

Now we are using two main board because of the limitation of our parts. In future work, we could purchase Grove Base Hat for Raspberry Pi and move all our sensors to Raspberry Pi instead of using two boards at the same time.

​

• Apply a facial detection to the image before we load the image into our model

Instead of current model we could can build a model similar to faster RCNN with an online image dataset on facial detection to determine of the face. Or we can use the YOLO model trained with annotated image dataset which can determine the boundary and class of image at the same time. However, the tradeoff between efficiency and accuracy of recognition is always the issue to discuss especially in such a wireless IoT system.