This GitHub repository contains a fork of the BFCM 2022 software with pre-given packages and examples demonstrating how to communicate with them. The pre-installed SD card is already configured to communicate with demo scripts within the ECC-BFCM/Computer project, and these scripts are set to run at start-up. However, if you wish to remove this configuration for development purposes, simply clean out the /etc/rc.local file.
In addition to the main version, the repository also includes custom ROS nodes for line detection, semaphore, person detection, and traffic sign detection added by our team.
Architecture
The implementation of the perception module is based on two ROS nodes, which can be found in the ./src/perception/src directory. The first node is responsible for publishing all the detected lanes, while the second node sends a message when a semaphore is detected along with its position.
Our code performs lane detection by applying a trapezoidal region of interest and a preprocessing pipeline that includes converting the image to grayscale, an edge detector, applying a Gaussian filter to reduce noise, and using a dilation filter to enhance edges. We then apply the Hough Transform algorithm to detect the lines in the image and calculate the average slope intercept to represent the road markings.
Additionally, our code contains a method called detect_crosswalk() that detects the presence of a crosswalk in an input image. This method creates a region of interest in the lower part of the image and detects contours using cv.findContours(). It then filters out small contours and checks if the number of remaining contours is greater than 2, indicating the presence of a crosswalk.
These methods can be used together to provide a more comprehensive understanding of the vehicle's environment.
To detect semaphores and other objects in the environment, we used the Yolo object detector. We were specifically interested in detecting the following object classes: person, bicycle, car, motorbike, aeroplane, bus, train, truck, traffic light, fire hydrant, stop sign, and parking meter.
The control module uses the information obtained from lane detection and object detection to make informed decisions about the vehicle's path planning and control.
The control module utilizes lane and object detection information to make informed decisions about the vehicle's path planning and control. A simple PID algorithm is implemented, with only the Kp parameter set. Additionally, to ensure smooth control of the vehicle's movement, the moving average method is applied to the last five series of lane positions. This approach helps to reduce the effect of sudden changes in lane detection and ensures more stable and predictable control of the vehicle.