Week 2 Journal

This week, Jennifer and I poured over more research papers to try to form a concrete plan for our project. One part of the project that was unclear was the camera needed to record the input. Dr. Briggs suggested using the OpenMV Cam, but since it can not be acquired, we have looked into many alternative options. As of now, the best option is to use the LCDK with a camcorder input.

In addition, we set basic milestones for our project. By week 3, we hope to have completed most of the preprocessing on the input. This will include using the OpenCV library to extract frames from the video, separate them into three separate channels (R, G, B), and then preprocess each frame using different kinds of filters. In addition, we will have to implement a face detection technique so we can isolate most of the background out of our algorithm. By week 7, we would like to have implemented the meat of the algorithm - finding the average intensity in each channel of each frame, and plotting it over time. For each signal (R, G, B), we would take the Fast Fourier Transform. The signal with the highest peak will be the one that most accurately represents the heart rate. Given that these milestones are completed on time, we can spend the rest of the quarter working on one or more stretch goals - making sure the algorithm works if more than one face is detected, comparing the efficiency of the algorithm in different light settings and skin tones. In addition, a major complication can arise from the complexity of the program. Since there are many calculations that need to occur in each frame, the complexity of the program will be dependent on the number of frames and therefore the length of the input. However, a certain length is required in order to have an accurate prediction, so finding a reasonable video length will be a challenge.

For now, the main priority is to make sure that the LCDK is able to read an input from the camcorder.