COMPUTER VISION & CONTROL PROJECTS
COMPUTER VISION & CONTROL PROJECTS
Embedded Files
Lane Detection Project
Lane Detection Project
In this project we attempt to create a system that determines driver intent when their vehicle begins leaving their initial lane on a highway. To do this, we detect the heading angle of the vehicle using one RGB camera located at the front, and from how the heading angle changes over time we can determine the driver’s intent.
We take stock dashcam footage from the internet where drivers are changing lanes normally vs falling asleep at the wheel and drifting. In order to calculate the heading angle of the car we need to find the relative angle of the lanes. After finding our region of interest we color threshold the image to only view standard white and yellow lane colors. We then perform a technique called sobel edge detection where we combine the x and y gradients of the image that we calculate using a kernel to find only where there are sharp changes in intensity. We can use the Hough Line detection algorithm to fit the pixels to best fit lines and use group those lines with K-means clustering to find the slope.
This process lets us calculate phi or heading angle of the car over time, and the way phi changes can allow us to determine whether or not the car's active lane control should take over. Read the full writeup my team did here:
https://drive.google.com/file/d/1uV28stvDRqtQW9xOKWoO4UuPHWoKsz57/view?usp=sharing
Inverse Pendulum Project
Inverse Pendulum Project
For this project, I created an inverse pendulum that maintained an upright position by spinning a motor at the top. The system uses conservation of angular momentum by spinning the motor in the opposite direction the arm wants to turn. To control the motor, I created a closed loop PID controller that determines how much the motor should spin.
For this project, I created an inverse pendulum that maintained an upright position by spinning a motor at the top. The system uses conservation of angular momentum by spinning the motor in the opposite direction the arm wants to turn. To control the motor, I created a closed loop PID controller that determines how much the motor should spin.
This project was mostly just practice to make use of control theory I learned in classes, which was successful because I learned a lot doing a practical problem. However, there are many real world cases where reaction wheels are used, such as many spacecraft and satellites. I implemented both PID and LQR control to compare the two.
Difficulties:
This project was mostly just practice to make use of control theory I learned in classes, which was successful because I learned a lot doing a practical problem. However, there are many real world cases where reaction wheels are used, such as many spacecraft and satellites. I implemented both PID and LQR control to compare the two.
Difficulties:
- The IMU has a little bit of gyroscopic drift, which causes the detected angle to shift over time. I've partially corrected for this with initialization code, but the system still ends up failing after a while.
- The system vibrates quite a bit, which throws off some of the readings sometimes. In the future I will print a thicker axle with tighter tolerances to reduce wobbling
- The connection between the motor and the wheel had some give, creating delay between when the motor spins and when the wheel spins. This was fixed by adding a set screw in between the motor and the wheel, however, the motor is using plastic gears which still has a little bit of the same problem.
Parts: LSMD60 IMU (accelerometer + gyroscope)
ESP32 Feather Dev Board
DRV8833 motor driver
9V Battery
Parts: LSMD60 IMU (accelerometer + gyroscope)
ESP32 Feather Dev Board
DRV8833 motor driver
9V Battery
Page updated
Report abuse