Evan Compton Status Report 5/4
This week as a group we finished integrating our parts and collected some test data for our wearable. We also prepared the slides for the presentation. Since then I have been trying to tweak things to improve the object recognition part of the project, as well as help out the other parts in any way I can. On Sunday we will do any last tweaking before the demo, and make the final video for our project.
Evan Compton Status Report 4/27
This week we tried to get all of our stuff running together integrated on the pi. The first issue that occurred is when running my cv stuff on the pi, we realized that it was very slow (despite being very fast on my computer), as it took 17 seconds on the pi and only 1 second on my laptop. So I re-wrote all of the cv stuff to be the exact same algorithm I had but very efficient (I had…
Evan Compton Status Report 4/20
This week I fixed up the car part of the algorithm to check multiple regions for of the cropped image for cars. Other than that I continued to tweak the initial cropping box parameters and add in online training data. I also helped Alli with getting her App on her phone. I’m not sure we will have time to collect a lot of training data with our wearable (It’s hard to do that without the other two parts of the…
Evan Compton Status Report 4/13
This week I added in car’s to my algorithm, which really pretty much completed the core tech of the object recognition part of the project (all that’s left is pretty much adding training data and tweaking the parameters). I also fixed a few bugs I found in some of the knn stuff when testing. Me and Alli also went around with our wearable and took a bunch of pictures in different locations of runners and cars, which has helped me…
Evan Compton Status Report 4/6
This week was spent getting the project far enough for the demo. I wrote the section picking algorithm for runners and biker’s, by implementing a knn influenced machine learning algorithm for picking the section most similar to the runner training images, and picking the section most similar to the biker images (total euclidean distance of closest k points), and going with the closer of the two. I also spent a good bit of time getting my stuff set up on…
Evan Compton Status Report 3/31
This week I came up with my initial approach for features for images (a bunch of blocks of percentage of key points of the images (i.e. 2×4 blocks). The exact number of blocks will be optimized once I have training/test data pictures. After collecting some photos with our camera at about 8m away, I have decided that only a certain section of the image will potentially contain the object we are trying to detect (in the middle of the image)…
Team update 3/16-3/23
This week we all continued to work individually on our parts of the project. No real changes to our design or schedule. Next week we would like to try to get all of our parts at least far enough for the demo and integrate what we have.
Evan Compton Status Report 3/16-3/23
This week I continued to work on and implement the image processing part of the project. I implemented the part to pick the section of the image the object we are trying to detect probably is in, based on which section has the most edges after blurring. I also implemented a k-nearest neighbors classifier. I have also put together the full workflow for the object recognition part (so the only parts that need to be improved are all of the…
Evan Compton Status Report 3/2-3/9
This week I did a lot of work on the object recognition part, to really explore the capabilities of OpenCV and finalize our design for this part of the project. As a group we also spent a lot of time doing the design report. I decided that the best way to do fast object recognition for this project is based on the geometrical differences between the 3 different types of objects. In experimenting with online images, I found that edge…