Vehicle Detection And Tracking

Self-Driving Car NanoDegree

Overview

In this project, our goal is to write a software pipeline to detect vehicles in a video (start with the test_video.mp4 and later implement on full project_video.mp4), but the main output or product we want to create is a detailed writeup of the project.

The goals / steps of this project are the following:

• Apply a color transform and append binned color features
• Perform a Histogram of Oriented Gradients (HOG) feature extraction on a labeled training set of images and train a classifier Linear SVM classifier
• Note: for those first two steps don't forget to normalize your features and randomize a selection for training and testing.
• Implement a sliding-window technique and use your trained classifier to search for vehicles in images.
• Run your pipeline on a video stream (start with the test_video.mp4 and later implement on full project_video.mp4) 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.

Here are links to the labeled data for vehicle and non-vehicle examples to train our classifier. These example images come from a combination of the GTI vehicle image database, the KITTI vision benchmark suite, and examples extracted from the project video itself.We can also take advantage of the recently released Udacity labeled dataset to augment our training data.

Writeup

Step 1: Color Histogram

Apply a color transform and append binned color features :

• Load the labeled data of vehichle and non-vehichle examples

• Choose one color space from RGB, HSV,LUV,HLS,YUV,YCrCb color space.
• Compute color histogram features
• Use cv2.resize().ravel() to create the feature vector

Step 2: Histogram of Oriented Gradients (HOG)

Perform a Histogram of Oriented Gradients (HOG) feature extraction on a labeled training set of images:

• Choose one color space from RGB, HSV,LUV,HLS,YUV,YCrCb color space.
• Separate color channels from color space
• Use skimage.feature.hog() to extract HOG feature
• Adjust the values of cells_per_block and orient(typical values are between 6 and 12 bins)

• Append color histogram features , spatial features and HOG features to features list
• Extract the feature vectors of vehichle and non-vehichle data set
• Split up data into randomized training and test sets
• Use sklearn.preprocessing.StandardScaler() to normalize the features,and apply the scaler to training and test data sets

Step 3: Building and Training Classifier

Use LinearSVC as the classifier :

Step 4: Sliding Window and search for vehicles

Implement a sliding-window technique and use your trained classifier to search for vehicles in images :

• Decide what size window you want to search
• Decide where in the image you want to start and stop your search
• Decide how much you want windows to overlap
• Get a list of windows to search for vehicles

• Iterate over all windows in the list
• Extract features for each window
• Scale extracted features
• Use trained classifier to predict
• Save the windows for positive detections

Step 5: Multiple Detections

• Use multi-scale windows to search for car
• Adjust the positions of start and stop to search

• Add heat to each box in box list,where a positive detection is predicted by SVC
• Apply threshold to remove false positives
• Use scipy.ndimage.measurements.label() to find final boxes from heatmap