This week we made strides in our feature extraction approach. Determining the location of the multimeter probe tip in a frame has proven to be a challenge. This is due to the camera perspective changing slightly during use. This issue is obvious in our newly collected dataset. Since finalizing the dataset, we have secured the camera mount on the probe to reduce this issue. Nonetheless, we have implemented template-matching in our feature extraction pin classifier to assist in probe tip localization. This works by taking a template of a portion of our probe, as seen above. Using this, the system attempts to match this template to a location in the image. From this location, we can deduce the position of the probe and tip.

As well, we have began our live testing process. A new script was created for this process of collecting a live testing dataset. This set will be used to finally validate our deep learning and feature extraction models.