Vehicle Detection and Tracking

Final Results via Deep Learning

Click to play video:

1st Implementation: Manual Feature Extraction + Classifier

Color Features
- Color Histogram
- Color Spaces (HSV, LUV, etc.)
- Spatial Binning (cv2.resize())
Gradient Features
- Histogram of Oriented Gradients (HOG)

Feature Extraction

Normalize each feature, then combine into one feature vector
Train-Test split
Could even use a decision tree for feature selection
Be careful of time dependencies, even w/ a random train-test split

Train Vehicle Classifier

Train SVM to classify Car/Not Car.

Sliding Window

Slide a window (at different scales) over the frames and for each window, classify car/not-car.

Manual Feature Extraction Results

Click to play video:

Deep Learning Implementation (YOLOv5)

Learning the features directly.

Click to play video: