Find the distance from the object to the camera using the YoloV4 object detector, here we will be using a single camera π·, detailed explanation of distance estimation is available in another repository Face detection and Distance Estimation using single camera
YoloV4.Distance.Estimation.mp4
Video Tutorial Explains the concept and implementation
-
Here we are targeting the person and cell phone classes only, for demo purposes.
-
you can follow all the steps mentioned in the video to create other objects as well.
implementation detail available on Darknet
- Finding the distance of multiple objects at the same time.
pip install opencv-contrib-python==4.5.3.56
pip3 install opencv-contrib-python==4.5.3.56
then just clone this repository and you are good to go.
I have used tiny weights, check out more on darknet GitHub for more
You will make changes on these particular lines DistanceEstimation.py
if classid ==0: # person class id
data_list.append([class_names[classid[0]], box[2], (box[0], box[1]-2)])
elif classid ==67: # cell phone
data_list.append([class_names[classid[0]], box[2], (box[0], box[1]-2)])
# Adding more classes for distance estimation
elif classid ==2: # car
data_list.append([class_names[classid[0]], box[2], (box[0], box[1]-2)])
elif classid ==15: # cat
data_list.append([class_names[classid[0]], box[2], (box[0], box[1]-2)])
# In that way you can include as many classes as you want
# returning list containing the object data.
return data_list
You have to make changes on these lines π DistanceEstimation.py
there are two situations, if the object(classes) in the single image then, here you can see my reference image
it has to two object, person and cell phone
# reading reference images
ref_person = cv.imread('ReferenceImages/image14.png')
ref_mobile = cv.imread('ReferenceImages/image4.png')
# calling the object detector function to get the width or height of the object
# getting pixel width for person
person_data = object_detector(ref_person)
person_width_in_rf = person_data[0][1]
# Getting pixel width for cell phone
mobile_data = object_detector(ref_mobile)
mobile_width_in_rf = mobile_data[1][1]
# getting pixel width for cat
cat_data = object_detector(ref_person)
cat_width_in_rf = person_data[2][1]
# Getting pixel width for car
car_data = object_detector(ref_person)
car_width_in_rf = person_data[3][1]
if there is single class(object) in reference image then you approach it that way π
# reading the reference image from dir
ref_person = cv.imread('ReferenceImages/person_ref_img.png')
ref_car = cv.imread('ReferenceImages/car_ref_img.png.png')
ref_cat = cv.imread('ReferenceImages/cat_ref_img.png')
ref_mobile = cv.imread('ReferenceImages/ref_cell_phone.png')
# Checking object detection on the reference image
# getting pixel width for person
person_data = object_detector(ref_person)
person_width_in_rf = person_data[0][1]
# Getting pixel width for cell phone
mobile_data = object_detector(ref_mobile)
mobile_width_in_rf = mobile_data[0][1]
# getting pixel width for cat
cat_data = object_detector(ref_cat)
cat_width_in_rf = person_data[0][1]
# Getting pixel width for car
car_data = object_detector(ref_car)
car_width_in_rf = person_data[0][1]
# Then you find the Focal length for each
If you have any doubt DM me on insta
if You found this Helpful, please star β it.
You can Watch my Video Tutorial on Computer Vision Topics, just check out my YouTube Channel
I am avalaible for paid work here Fiverr