Week 11 Journal

The main focus of this week was to add improvements to the heart rate detection algorithm, and work on a couple stretch goals.

During our research, Jennifer and I saw that face detection may be more time consuming than face tracking. In order for face tracking to work well, a face must be detected at certain intervals. The keypoints of the detected face will be saved, and then every time a resemblance of a face is seen in the region of interest, the keypoints of this are matched with the saved keypoints in order to determine a match. Therefore, instead of trying to detect a face every frame, we are trying to match keypoints. While this does not lessen the complexity of the algorithm, it allows us to get more data points because tracking a face is faster than detecting one. Also, tracking a face allows us to make sure that the face detected is the same one throughout the data collecting process.

In addition to implementing the face tracking, we experimented with the window size and the threshold parameters of the face detection. This led to better results because more of the frame was devoted to the face, so the forehead region had more pixels to extrapolate data from.

Finally, we started to clean up the code and noticed that removing print statements (that were used for debugging purposes) allowed the algorithm to collect even more data. Before getting rid of the print statements, the frames per second would start at 16 fps and decrease to 12 fps near the end of the data collection. After cleaning up a portion of the code, the algorithm started at 17 fps and decreased down to 15 fps. This allowed us to receive more data, which should help us get a more accurate heart rate.

Next week, we plan on trying a couple different test cases with the algorithm - maybe test of filters and different types of lighting. We started to run trails on different amount of face detection and made tables to show the average error of the algorithm. Hopefully, we will continue to run more trials and work on the report and presentation.