In Project 5 of the great Udacity Self Driving car nanodegree, the goal is to use computer vision techniques to detect vehicles in a road.
The project is completed in the following stages:
- Step 1: Create a function to draw bounding rectangles based on detection of vehicles.
- Step 2: Create a function to compute Histogram of Oriented Gradients on image dataset.
- Step 3: Extract HOG features from training images and build car and non-car datasets to train a classifier.
- Step 4: Train a classifier to identify images of vehicles.
- Step 5: Identify vehicles within images of highway driving.
- Step 6: Track images across frames in a video stream.
HOG stands for “Histogram of Oriented Gradients”. Basically, it divides an image in several pieces. For each piece, it calculates the gradient of variation in a given number of orientations. Example of HOG detector — the idea is the image on the right to capture the essence of original image HOG will compute the gradients from blocks of cells. Then, a histogram is constructed with these gradient values.
def get_hog_features(img, orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True):
# Call with two outputs if vis==True
if vis == True:
features, hog_image = hog(img, orientations=orient, pixels_per_cell=(pix_per_cell, pix_per_cell),
cells_per_block=(cell_per_block, cell_per_block), transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features, hog_image
# Otherwise call with one output
else:
features = hog(img, orientations=orient, pixels_per_cell=(pix_per_cell, pix_per_cell),
cells_per_block=(cell_per_block, cell_per_block), transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features
For this step I defined a function get_hog_features which uses the Scikit-image function hog() to get Histogram of Oriented features from an image. The functions computes the HOG features from each of the 3 color channels in an image and concatenates them to gather in to a single feature vector.
def extract_features(imgs, cspace='RGB', orient=8,
pix_per_cell=8, cell_per_block=4, hog_channel=0):
# Create a list to append feature vectors to
features = []
# Iterate through the list of images
for file in imgs:
# Read in each one by one
image = mpimg.imread(file)
# apply color conversion if other than 'RGB'
if cspace != 'RGB':
if cspace == 'HSV':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)
elif cspace == 'LUV':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2LUV)
elif cspace == 'HLS':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2HLS)
elif cspace == 'YUV':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2YUV)
elif cspace == 'YCrCb':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2YCrCb)
else: feature_image = np.copy(image)
# Call get_hog_features() with vis=False, feature_vec=True
if hog_channel == 'ALL':
hog_features = []
for channel in range(feature_image.shape[2]):
hog_features.append(get_hog_features(feature_image[:,:,channel],
orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True))
hog_features = np.ravel(hog_features)
else:
hog_features = get_hog_features(feature_image[:,:,hog_channel], orient,
pix_per_cell, cell_per_block, vis=False, feature_vec=True)
# Append the new feature vector to the features list
features.append(hog_features)
# Return list of feature vectors
return features
After extracting features from all data, I used a Support Vector Machine Classifier (SVC), with linear kernel, based on function SVM from scikit-learn. In the end, I used color space YCrCb, all channels.
Before training the data, the data was normalized using StandardScaler() from sklearn.preprocessing.
Then these normalized data were splitted into train and test sets.
- HOG parameter: orient = 8, pix_per_cell = 8, cell_per_block = 2
svc.fit(X_train, y_train)
HOG Subsampling is used to have a efficient way to identifying hog feature in one go. After trying several different scale for different window sizes, I realised using only one scale would be a lot more effective as it allows less false positives.
def find_cars(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins):
boxes =[]
draw_img = np.copy(img)
img = img.astype(np.float32)/255
img_tosearch = img[ystart:ystop,:,:]
ctrans_tosearch = convert_color(img_tosearch, conv='RGB2YCrCb')
if scale != 1:
imshape = ctrans_tosearch.shape
ctrans_tosearch = cv2.resize(ctrans_tosearch, (np.int(imshape[1]/scale),
np.int(imshape[0]/scale)))
ch1 = ctrans_tosearch[:,:,0]
ch2 = ctrans_tosearch[:,:,1]
ch3 = ctrans_tosearch[:,:,2]
# Define blocks and steps as above
nxblocks = (ch1.shape[1] // pix_per_cell)-1
nyblocks = (ch1.shape[0] // pix_per_cell)-1
nfeat_per_block = orient*cell_per_block**2
# 64 was the orginal sampling rate, with 8 cells and 8 pix per cell
window = 64
nblocks_per_window = (window // pix_per_cell)-1
cells_per_step = 2 # Instead of overlap, define how many cells to step
nxsteps = (nxblocks - nblocks_per_window) // cells_per_step
nysteps = (nyblocks - nblocks_per_window) // cells_per_step
# Compute individual channel HOG features for the entire image
hog1 = get_hog_features(ch1, orient, pix_per_cell, cell_per_block, feature_vec=False)
hog2 = get_hog_features(ch2, orient, pix_per_cell, cell_per_block, feature_vec=False)
hog3 = get_hog_features(ch3, orient, pix_per_cell, cell_per_block, feature_vec=False)
for xb in range(nxsteps):
for yb in range(nysteps):
ypos = yb*cells_per_step
xpos = xb*cells_per_step
# Extract HOG for this patch
hog_feat1 = hog1[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_feat2 = hog2[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_feat3 = hog3[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3)).reshape(1, -1)
xleft = xpos*pix_per_cell
ytop = ypos*pix_per_cell
# Scale features and make a prediction
test_features = X_scaler.transform(hog_features)
#test_features = X_scaler.transform(np.hstack((shape_feat, hist_feat)).reshape(1, -1))
test_prediction = svc.predict(test_features)
if test_prediction == 1:
xbox_left = np.int(xleft*scale)
ytop_draw = np.int(ytop*scale)
win_draw = np.int(window*scale)
cv2.rectangle(draw_img,(xbox_left, ytop_draw+ystart),(xbox_left+win_draw,ytop_draw+win_draw+ystart),
(0,0,255),6)
boxes.append(((xbox_left, ytop_draw+ystart),(xbox_left+win_draw,ytop_draw+win_draw+ystart)))
return draw_img, boxes
For this step I defined a function draw_boxes which takes as input a list of bounding rectangle coordinates and uses the OpenCV function cv2.rectangle() to draw the bounding rectangles on an image.
def add_heat(heatmap, bbox_list):
# Iterate through list of bboxes
for box in bbox_list:
# Add += 1 for all pixels inside each bbox
# Assuming each "box" takes the form ((x1, y1), (x2, y2))
heatmap[box[0][1]:box[1][1], box[0][0]:box[1][0]] += 1
# Return updated heatmap
return heatmap# Iterate through list of bboxes
def apply_threshold(heatmap, threshold):
# Zero out pixels below the threshold
heatmap[heatmap <= threshold] = 0
# Return thresholded map
return heatmap
For the video implementation, I have used a weighted averaged heatmap to reduce false postive and increase robustness of the pipeline. Specificlly, I do a weight average of the 4 frames, with high importance of the most recent frame, and used that as the heatmap instead.
heat1 = np.zeros_like(img[:,:,0]).astype(np.float)
heat2 = np.zeros_like(img[:,:,0]).astype(np.float)
heat3 = np.zeros_like(img[:,:,0]).astype(np.float)
def process_image(img):
global heat1
global heat2
global heat3
global heat4
ystart = 400
ystop = 656
scale = 1.35
colorspace = 'YCrCb' # Can be RGB, HSV, LUV, HLS, YUV, YCrCb
orient = 8
pix_per_cell = 8
cell_per_block = 2
hog_channel = 'ALL' # Can be 0, 1, 2, or "ALL"
spatial_size = (16, 16)
hist_bins = 32
spatial_feat = True
hist_feat = True
hog_feat = True
svc = pickle.load( open("saved_svc.p", "rb" ) )
X_scaler =pickle.load( open( "saved_X_scaler.p", "rb" ) )
out_img, boxes = find_cars(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell,
cell_per_block, spatial_size, hist_bins)
heat = np.zeros_like(img[:,:,0]).astype(np.float)
heat = add_heat(heat,boxes)
heat = 0.7*heat + 0.25*heat1+ 0.05*heat2
heat = apply_threshold(heat,1.5)
heatmap = np.clip(heat, 0, 255)
labels = label(heatmap)
result=draw_labeled_bboxes(img, labels)
heat1=heat
heat2=heat1
return result
Video link: https://youtu.be/9z2tveQMJk0
The hard part of the project is the elimination of false positives. The pipeline as it is still falsely identify road/tree/etc as a car. The most efficient way to improve this model is to use Deep learning like SSD or training a more robust identifier (99%).
Additionally, the pipeline is running very slowly, at serveral seconds per frame. Moving froward, it would be needed to process the image in real time using Deep learning or pruning the number of features and reducing the number of windows searched.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent