Vehicle Detection and Classification

A project by Tanmaya Dabral, Bhargav Kanuparthi and Sabrinath Harilal as a part of the Image Processing course (BITS F311) offered at BITS Pilani, Hyderabad Campus. The aim of the project was to detect and classify vehicles from a video of an Indian highway, the classification labels being "Heavy" and "Light". Both supervised and unsupervised approaches were tried.

Note: Since committing thousands of frames every time they are processed is a hassle, only zipped versions are committed.

Method used

The vehicle blobs were first obtained by subtracting the images from the median image followed by elementary cleaning operations. For the unsupervised approach, the blobs were clustered based on their area and their median width to median height ratio. For the supervised approach, a few high-precision low recall samples of both classes were obtained using blob statistics. Then, a Haar cascade classifier was trained for each class using these training examples.

Results

The unsupervised clustering performs reasonably well on both classes. However, the supervised approach gives far too many false negatives for the heavy class to be acceptable. The "Light" class however, is predicted with greater accuracy by the Haar cascade classifier.

Clustering examples:

Haar cascades examples (Two-Wheelers):

Haar cascades examples (Four-Wheelers)::

Vehicle classification

Bro i want to classify vehicles but unable to do so.this is my code plz help me in classification as light and heavy vehicles.plz help me bro....m

import cv2

import numpy as np

min_contour_width=40 #40

min_contour_height=40 #40

offset=10 #10

line_height=550 #550

matches =[]

vehicles=0

def get_centroid(x, y, w, h):

x1 = int(w / 2)

y1 = int(h / 2)

cx = x + x1

cy = y + y1

return cx,cy

#return (cx, cy)

#cap = cv2.VideoCapture(0)

cap = cv2.VideoCapture('video.mp4')

#cap = cv2.VideoCapture('Relaxing highway traffic.mp4')

cap.set(3,1920)

cap.set(4,1080)

if cap.isOpened():

ret,frame1 = cap.read()

else:

ret = False

ret,frame1 = cap.read()

ret,frame2 = cap.read()

while ret:

d = cv2.absdiff(frame1,frame2)

grey = cv2.cvtColor(d,cv2.COLOR_BGR2GRAY)

#blur = cv2.GaussianBlur(grey,(5,5),0)

blur = cv2.GaussianBlur(grey,(5,5),0)

#ret , th = cv2.threshold(blur,20,255,cv2.THRESH_BINARY)

ret , th = cv2.threshold(blur,20,255,cv2.THRESH_BINARY)

dilated = cv2.dilate(th,np.ones((3,3)))

kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (2, 2))

# Fill any small holes

closing = cv2.morphologyEx(dilated, cv2.MORPH_CLOSE, kernel)

contours,h = cv2.findContours(closing,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)

for(i,c) in enumerate(contours):

(x,y,w,h) = cv2.boundingRect(c)

contour_valid = (w >= min_contour_width) and (

h >= min_contour_height)

if not contour_valid:

continue

cv2.rectangle(frame1,(x-10,y-10),(x+w+10,y+h+10),(255,0,0),2)

cv2.line(frame1, (0, line_height), (1200, line_height), (0,255,0), 2)

centroid = get_centroid(x, y, w, h)

matches.append(centroid)

cv2.circle(frame1,centroid, 5, (0,255,0), -1)

cx,cy= get_centroid(x, y, w, h)

for (x,y) in matches:

if y<(line_height+offset) and y>(line_height-offset):

vehicles=vehicles+1

matches.remove((x,y))

print(vehicles)

cv2.putText(frame1, "Total vehicles Detected: " + str(vehicles), (10, 90), cv2.FONT_HERSHEY_SIMPLEX, 1,

(0, 170, 0), 2)

#cv2.drawContours(frame1,contours,-1,(0,0,255),2)

cv2.imshow("Original" , frame1)

#cv2.imshow("Difference" , th)

if cv2.waitKey(1) == 27:

break

frame1 = frame2

ret , frame2 = cap.read()

#print(matches)

cv2.destroyAllWindows()

cap.release()

many-facedgod / vehicle-detection-and-classification Goto Github PK