xiaolaozai / car-detector

HOG + SVM classifier trained to detect cars

#car-detector Uses histogram of oriented gradients as a descriptor and a linear support vector machine as a classifier.

Uses non-maximum suppression to eliminate overlapping bounding boxes.

#####Sample

#####Usage python car_detector/detect.py -i data/sample/test-101.jpg

#####Files

• config.py contains general configuration values
• detect.py runs classifier and draws bounding boxes on sample image
• extract-features.py converts images into HOG descriptors and writes to disk to be loaded during training
• helpers.py contains various helper functions
• test.py tests classifier on training or test set printing precision and recall
• train.py trains classifier on files output by extract-features

#####Limitations

• Only detects cars from profile view
• Has trouble with larger images

#####Resources

• HOG+SVM Guide
• NMS
• Training data