Project

# Title Team Members TA Documents Sponsor
76 Driver Fatigue System
 Julio Cornejo
Vincent Ng
 Maanas Sandeep Agrawal design_document1.pdf
final_paper1.pdf
proposal1.pdf
video
Driver Fatigue System

Team Members:
* Julio Cornejo (jcorne23)
* Vincent Ng (vng20)

**Problem**

When driving for prolonged amounts of time, some key body movements and facial changes can be made due to drowsiness. The drowsiness, if unmonitored, can pose dangerous conditions for other drivers and the drivers themselves. Intoxication while driving is also a rampant issue; there is no universal breathalyzer that prohibits driving based on BAC. I propose this device that uses facial recognition and eye-level detection sensors and cameras to detect symptoms of fatigue along the road while also prohibiting intoxicated drivers from proceeding to drive. This device can monitor head position, yawns, and register long blinks. It can also track the driving duration and eventually register all these symptoms if it detects fatigue. Once enough triggers are set, an app interface can assess your tiredness or driving incapability via Wifi transmission. It can suggest and locate the nearest rest stop or call emergency contacts (set by the user). When certain drowsiness scores are reached, the user's BAC and live drowsiness rating will be displayed with in-house buzzer systems.

**Solution**

The system revolves around an algorithm that makes use of a variety of sensors and cameras. One is a breathalyzer that measures the blood alcohol concentration of the driver. Using the software, we monitor the live value and set triggers to call emergency contacts and monitor until a safe concentration for resumable driving. Drowsiness and tiredness can be detected in various ways ranging from yawn frequency, long blinks, and head tilts. They happen suddenly. Most traditional cars use your position in a lane to track tiredness. Tracking head movement and analyzing the face for more key indicators precisely provides more information that can be essential to identify when a driver needs to step away from the wheel and request a ride elsewhere. The PCB can be housed in a small, compact shape like a cube that sits over any dashboard with detachment features to trigger the BAC sensor correctly.

**Solution Components:**

**PCB Controller:**

The device will have operable buttons to measure and clear the sensor and turn on the components and precise data for use. ESP32-S3 is the microcontroller intended for this project since the WiFi connection is crucial.

**Sensor and Device Subsystem**

* OV2640 Camera Module (For eye & head movement detection)
* BNO055 IMU (For head tilt & movement tracking)
* DFRobot Gravity MQ-3 Sensor (For BAC detection)
* OLED DISPLAY (e.g., SSD1306) displays live-drowsiness score calculation and BAC reads.
* BUZZER (if there is a significant change in the drowsiness measure, we ring a buzzer for the first tier of detections).

**Power Subsystem**

Power Management (Li-ion battery) - A rechargeable lithium-ion battery for a recommended runtime of about 10 hours would suffice for long drives.

**Algorithm**

The most ethic-heavy aspect of this project is to design a rationale and fair approach to assess and set triggers for the alerts. Setting criteria for calculating a drowsiness score for periodic yawns, blink duration, and rampant head movements. Not to mention, algorithmically, we have to assess what constraints from the sensors classify a yawn or blink, which can be difficult depending on the light settings. The goal is to be processed within the ESP32-S3 as the algorithm's in-house computation, with the possibility of an out-of-PCB Raspberry PI for additional computation power (if necessary).
To capture blink frequency/length, we will be utilizing Haar cascades, or some sort of pretrained model if we decide to implement the raspberry pi for processing. For drowsiness detection, we will be implementing an Eye Aspect Ratio algorithm to check if the drivers eyes are closed and blink frequency/length will be tracked based on how many frames are in the eye closed EAR threshold and how many frames are in the eyes open EAR threshold.


**Sample Weighted Calculation (idea):**

Fatigue Score = (Weight1 * Blink Score) + (Weight2 * Yawn Score) + (Weight3 * Head Movement Score)
For example, Blink Score could be calculated as:
* Blink frequency (e.g., 0-5 blinks/min → score 0-50).
* Blink length (e.g., >500ms → score 0-50).
* Final Drowsiness Level:
- Score 0-30%: Alert (Buzzer or immediate audio alert)
- Score 30-60%: Warning
- Score 60-100%: Fatigued (Follow user designated-procedure)


**User Interface & Connectivity**

The PCB would send data via WIFI from the microcontroller through the ESP-32. Using HTML and CSS, we can design a mobile application that includes charts and emergency functionality to follow procedures like assigning and calling emergency contacts once a drowsiness or BAC threshold is reached. The focus is on a mobile application, as a driver usually has that in hand.

**Criterion for Success**

1. Accurate measurement of BAC levels: The device should be able to measure the BAC levels of the driver accurately. Having a professional police-enforced measure for comparison can help calibrate.

2. Effective detection of yawns, blinks, and tilting heads: The device should be able to classify and count how many yawns and blinks occur in timed driving intervals while actively updating and displaying the tiredness score for the driver to be wary. We can test this with varying yawn lengths and close-eye durations.

3. Fair Algorithm: The design is intended to save lives. We want the user to be able to set up a procedure when a threshold is set up. This can be calling driving services or an emergency contact once a BAC exceeds this.

4. Updated interface: The web interface should reflect updates and show graphic analytics to predict better when tiredness occurs and optimal driving times for all safety.

GYMplement

Srinija Kakumanu, Justin Naal, Danny Rymut

Featured Project

**Problem:** When working out at home, without a trainer, it’s hard to maintain good form. Working out without good form over time can lead to injury and strain.

**Solution:** A mat to use during at-home workouts that will give feedback on your form while you're performing a variety of bodyweight exercises (multiple pushup variations, squats, lunges,) by analyzing pressure distributions and placement.

**Solution Components:**

**Subsystem 1: Mat**

- This will be built using Velostat.

- The mat will receive pressure inputs from the user.

- Velostat is able to measure pressure because it is a piezoresistive material and the more it is compressed the lower the resistance becomes. By tracking pressure distribution it will be able to analyze certain aspects of the form and provide feedback.

- Additionally, it can assist in tracking reps for certain exercises.

- The mat would also use an ultrasonic range sensor. This would be used to track reps for exercises, such as pushups and squats, where the pressure placement on the mat may not change making it difficult for the pressure sensors to track.

- The mat will not be big enough to put both feet and hands on it. Instead when you are doing pushups you would just be putting your hands on it

**Subsystem 2: Power**

- Use a portable battery back to power the mat and data transmitter subsystems.

**Subsystem 3: Data transmitter**

- Information collected from the pressure sensors in the mat will be sent to the mobile app via Bluetooth. The data will be sent to the user’s phone so that we can help the user see if the exercise is being performed safely and correctly.

**Subsystem 4: Mobile App**

- When the user first gets the mat they will be asked to perform all the supported exercises and put it their height and weight in order to calibrate the mat.

- This is where the user would build their circuit of exercises and see feedback on their performance.

- How pressure will indicate good/bad form: in the case of squats, there would be two nonzero pressure readings and if the readings are not identical then we know the user is putting too much weight on one side. This indicates bad form. We will use similar comparisons for other moves

- The most important functions of this subsystem are to store the calibration data, give the user the ability to look at their performances, build out exercise circuits and set/get reminders to work out

**Criterion for Success**

- User Interface is clear and easy to use.

- Be able to accurately and consistently track the repetitions of each exercise.

- Sensors provide data that is detailed/accurate enough to create beneficial feedback for the user

**Challenges**

- Designing a circuit using velostat will be challenging because there are limited resources available that provide instruction on how to use it.

- We must also design a custom PCB that is able to store the sensor readings and transmit the data to the phone.