Category: Progress Report

This week I added the Manual Adjustment page to the GUI which will allow users to adjust both angle and height of the stand using button clicks that send serial commands to the Arduino similar to the system used for automatic adjustment.

I also created a focus tracking program using OpenCV that tracks the direction in which a user is looking to determine whether they are looking at their screen. This program keeps track of variations of both horizontal and vertical distances between key facial landmarks to periodically check the user’s head position, and if they have been consistently looking forward at the screen for 20 minutes, not including small breaks or movements, then they will receive a notification in the GUI reminding them to take a break to prevent eye strain. I also wrote a similar, simpler program that periodically detects whether a face is detected in frame which will be used to send notifications via GUI when a user should get up to stretch after long periods of sitting in front of their screen.

I was also able to begin testing the automated adjustment process using the linear actuators on our stand and adjusting the height of the stand manually. I tested the angle adjustment by opening my laptop to several different angles, undergoing calibration, running the angle adjustment code in the integrated subsystem, and tuning the time needed between feedback checks in OpenCV/Python. I found that 250 ms is the smallest delay that can be used between checking landmark distances before the angle adjustment becomes noticeably inaccurate.

I am currently on track, and next week, I plan to integrate the focus tracking aspects to the GUI. After we are able to supply enough torque for the platform jack’s screw rotation, I will more thoroughly test the height adjustment with the automated system.

New Tools and Knowledge:

Over the course of this project, I learned how to use OpenCV and other computer vision tools such as trained models for object classification and detection. To learn about these topics, I used many online tutorials, videos, and the OpenCV documentation to learn the basics of CV programs, and I wrote a bunch of smaller test programs to test out and learn about new features and how I could use them in our project.

This week I continued working on the GUI and added additional pages and features. I added a new page that allows a user to choose whether to calibrate their stand for the first time or to adjust the stand based on prior calibration data. This way, users can skip the calibration after they’ve gone through the process once. I was able to integrate the height and angle adjustment program with the Arduino via serial connection, and we are able to automate the motor and linear actuator movement.

I have done some testing with the height and angle adjustment to make sure that the algorithm we use is able to determine the correct height and angle for a user when starting from different heights and angles. I will continue testing this when we integrate the software with the stand in order to tune and make some adjustments to the time between feedback checks and speed at which we turn the motors in order to reach the ideal height and angle within our desired time limit.

I am currently on schedule. Next week I plan to add more pages to the GUI to display posture tracking data as well as implement eye tracking to the software in order to send users notifications when they have been focusing on their screen for long periods of time.

This week I wrote the Python code that would carry out the height and angle determination of our system. The process will first need a calibration stage in which the user’s landmark data will be captured when their laptop is at the ideal height. Then, any time the user uses the device again, this data will be used as reference for determining whether to increase or decrease the height and/or angle of the laptop. The program uses an iterative process of incrementally increasing or decreasing the angle of the laptop until the vertical distance between landmarks is approximately the same as the reference distance from calibration, and then switching to adjusting the height of the laptop until the y-coordinate of the landmark at the center of the face is within threshold of the position determined at calibration. This method proved to be pretty accurate even when starting at different heights and angles.

I am also working on a method that uses position of key landmarks on the frame’s coordinate system in order to determine the height and angle adjustments that will be needed. In this method, we wouldn’t need a calibration stage for the user which may be more convenient and faster, but it seems to be less accurate in my testing.

I am currently on schedule and will continue working on this software next week and am planning to have this height and angle adjustment code integrated with the Arduino firmware in order to move the motor and linear actuators according to which adjustment is needed.

This week I continued working on our automatic height and angle adjustment software. After discussing with the team, we decided to do a multiple iterations of height and angle adjustment to incrementally reach the correct position rather than calculate the exact height and angle needed for adjustment. The method I am considering is to use both the vertical distance between landmarks and the y-coordinate of the facial landmark in the middle of the user’s face to determine whether the screen is positioned at the correct height and angle. Using these two measurements, we can switch between incrementing height or angle by an interval and checking these measurements until we reach an acceptable threshold of vertical landmark distance and position of the face on the screen.

I also worked on creating the GUI for our system and added the front page with buttons to choose between opening a window to use the adjustment functionality or to view the posture tracking data. When a user clicks the adjustment button, the calibration and height determination program will run.

I am currently on schedule and am aiming to finish the iterative adjustment cycle and have it coded and tested by next week.

This week, I worked on a function that get the real life distance a laptop would need to rise based on the distance in pixels. To do this, I’m using the camera calibration as described in the OpenCV documentation to find the focal length of my camera and used these in my calculations assuming the distance from the screen is constant.

I also integrated the calibration setup for gathering facial landmarks with the current landmark distance calculation, so that we are using the calibration data points to calculate the reference difference and compare them to current landmarks. I’m able to detect if the camera is at a height lower than the one determined at calibration, higher, or within threshold.

However, when neither the height or pitch of the laptop are known, it is difficult to determine whether the correct course of action for the system will be to raise the height of the stand or to angle to the laptop using the linear actuators since either a change in height or change in angle can produce similar results in vertical landmark difference. In order to solve this issue, I’m working on using a few vectors between landmarks to measure the skew of certain sections of a user’s face to be able to decide if the camera is at an angle or at a height difference.

I am currently on schedule, and next week I will continue working on this height vs. angle determination. I also hope to begin creating a starter GUI to display basic functionalities such as calibration and setup.