Traffic Light Detection

Traffic light detection for autonomous driving

For detailed explanations visit medium.com/@kenan.r.alkiek and docs/

Dependencies

cv2
numpy
sklearn
skimage
matplotlib
yaml

Sample Usage

Train A Model

Customize settings in config.yaml (see config options for more information)
Run train.py

Run A Model

Customize settings in config.yaml (see config options for more information)
Run run.py

Config Options

Train Options

descriptor - str: hog, lbp, or haar
classifier - str: svm (more options to be added)
outfile - str: File location to save the trained model to
positive_image_directory - str: Where the positive images are stored
negative_image_directory - str: Where the negative images are stored
window_size - tuple (int, int): Size of the training windows

Run Options

descriptor - str: hog, lbp, or haar
classifier_location - str: Location of the saved classifier (the output of running train.py)
detector - str: spotlight, color
heatmap_memory - int: Number of frames to retain before removing the first frame from memory
heatmap_threshold - int: Number of overlapping detections in the heatmap before accepting a detection
window_size - tuple (int, int): Size of the boxes to classify and draw on
image_directory - str: Image directory to run the classifier on

Descriptor Options

HOG

block_size - tuple (int, int): Number of cells in each block
cell_size - tuple (int, int): Size (in pixels) of a cell
orientations - int: Number of orientation bins

LBP

points - int: Number of circularly symmetric neighbour set points
radius - float: Radius of circle (spatial resolution of the operator)

Haar

selected_feature_file - str: File containing the coordinates and types of the Haar features you want to use. More information in the docs

Detector Options

Spotlight

max_size - int: How large a spotlight can grow before being rejected
kernel - int: Size of the kernel to apply during top-hat morphology
threshold - int: Threshold value applied after top-hat morphology