Author: rngo500

Unfortunately, Jasper was out this week due to covid, but with his input the integration between the UI and the Computer Vision, as well as the integration between the Computer Vision and the Cooking Timer, is complete. The progress we hope to make was delayed several days for integration because the Xaiver broke halfway into the week, requiring several days and a reflash to fix. As a result of that, some progress was lost, but that loss in progress has been reversed with more work.

Furthermore, testing has begun for my computer vision modules for the final presentation. Most of the metrics seem to be fine, however the edge detection might be a bit off in terms of its predictions, probably due to camera quality factors that are hard to resolve. This is unfortunately one tradeoff that cannot be avoided since OpenCV has a max resolution, and any increases in resolution for the image severly impacts negatively everything else in the cv pipeline. Solution is to increase brightness even more, since there is only so much an edge detection algorithm can do when noise is present.

I updated my dataset with more images, including hard to find images of arms over the grill, in response to the feedback from the ethics class about potential harm from the project. The newly created network works. Metrics might need to be shifted a bit in the final report, however, since the categories for anything that is not meat is so broad and the dataset is so limited that there will be limits in terms of accuracy. I will see what I can do.

 

Jasper and I got the ethernet working, however, with the revelation that the final demo will take place in the UC, orders have been placed for both a router and a wifi card. Both orders placed because flexibility in terms of results.

 

I have begun integration of the cooking time algorithm with computer vision. I will need Jasper to integrate the user interface. Unfortunately, Joseph’s lack of completion of work in time for integration has bottlenecked progress and he will need to step up for my part of the scheduled tasks to be completed.

In the days before the demo, I modified the pixel to inch parameter to fit the size of the meat within the range of error. This value is hardcoded instead of dynamically determined because the meat is anticipated to be a fixed distance from the camera in the final product, so having fixed values instead of a reference point is better.

Before the demo, I was able to download the necessary libraries and files onto the Jetson, however I lacked enough adaptors to get the keyboard and the camera both working at the same time, so I was unable to get that working before the demo. I was able to get the necessary adaptors to get the Computer Vision working on the Jetson during Carnvial.

In the previous week, I combined the 3 different computer vision algorithms into one program that invokes all of them. Along the way, I found out the blob detection was not as effective as I thought under dim lighting conditions. As a result, I had to make modifications to the blob detection by changing several parameters regarding eroding excess lines.

You may have noticed the image is a result of an object recognition and not of an image cassification network. There are several reasons for this. One, the dataset we collected had multiple types of meat strewn together, and at that moment, our team realized that if someone placed multiple types of meat on a plate in front of the robotic arm, a classification system is not robust enough to know that and may cause undercookng of meat. Another reason is that if the blob detection fails to work in a way we want, object recognition algorithm is our backup mitigation technique. While object recognition is slower than blob detection, it’s still sufficiently fast enough for our desired metrics.

What you see below is the result of YOLOv5 trained on 150 epochs on a tiny data set (only 20 images) augmented to be 60 images in total. The batch size each epoch was 12 images each. YOLOv5 was selected due to its speed advantage and its active community support online.

Currently on track to complete by the completion date indicated on the schedule, which is Monday, at least I am mostly done. The hesitation is because the integration period will provide a chance to add to the dataset, and that would require more training on the network.

By next week, I hope to begin the integration of the subsystems by having the files uploaded onto the Xavier. Hopefully that would lead to an improvement in detection time also.

As promised last week, I was able to solve the issue of blob detection not working on larger blobs. As it turns out, the reason why I was unable to solve the issue earlier was because I was turning off filtering mechanisms one by one, however, I actually had to disable multiple preset parameters to get the blob detection to work on larger images. Furthermore, this past week I was working on ways to improve the blob detector and edge detector on different types of images. As seen in last week’s report, back then I was only testing on a simple slab of meat on a white background. This week, I obtained more images, especially the type of image seen below where the meat is on a tray, which is how I expect the final environment to work. And as you can see, my changes in parameters means edge and blob detection work in more scenarios accurately (note the 3 circles for the 3 trays of meat).

Next week, my 2 primary tasks involve beginning the setup of a neural network and getting the blob and edge detectors to take in a live camera feed to test its effectiveness in a more real world setting. I anticipate the collection and tagging of a dataset.

I was able to modify test images to isolate red portions of the image. From there, I was also able to easily obtain the outer edges using the canny edge detector. Isolating red portions of the image is necessary for the development of the computer vision system because the meat is red and the canny edge detection on an unmodified image does not return a clean outer border around the meat. (shown below)

The main issue I need to get solved by next week to remain on track is to figure out why the blob detector cv function does not work well with the processed binary image (shown below). Otherwise, I am on track to complete a significant portion of the width detection and blob detection algorithms.

The reason preprocessing needs to be done on the image is that meat usually is not on a white background like this. Therefore, necessary distractions must be removed for a more accurate detection of the material.

By next week, some form of blob detection must work on this test image, and ideally one other more crowded image.