Vehicle Detection for Self-Driving Cars
The objective of this project is to create a pipeline (model) to draw bounding boxes around cars in a video. Example images and video come from a combination of the GTI vehicle image database and the KITTI vision benchmark suite. We have the following goals:
- Perform a Histogram of Oriented Gradients (HOG) feature extraction on a labeled training set of images and train a classifier Linear SVM classifier
- Implement a sliding-window technique and use trained classifier to search for vehicles in images
- Run pipeline on a video stream and create a heat map of recurring detections frame by frame to reject outliers and follow detected vehicles
- Estimate a bounding box for vehicles detected
First we perform a Histogram of Oriented Gradients (HOG) feature extraction on a labeled training set of images.
Here are examples of images from the vehicle class and the non-vehicle class.
Here are some visualized HOG features extracted from the images.
We can implement a sliding-window technique and use our trained classifier to find vehicles in images.
Initially using HOG features and SVM, a heatmap could average the detection results from successive frames. The heatmap was thresholded to a minimum value before labeling regions, to remove false positive. This process was shown above.
With deep learning, however, we can rely on a confidence score to decide the tradeoff between precision and recall. This next figure shows the effect of thresholding SSD detection at different level of confidence.