Yolov4-Detector-and-Distance-Estimator

Find the distance from the object to the camera using the YoloV4 object detector, here we will be using a single camera :camera:, detailed explanation of distance estimation is available in another repository Face detection and Distance Estimation using single camera

https://user-images.githubusercontent.com/66181793/124917186-f5066b00-e00c-11eb-93cf-24d84e9c2e7a.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

TO DO

Finding the distance of multiple objects at the same time.

Installation you need opencv-contrib-python

opencv contrib

windows

pip
pip install opencv-contrib-python==4.5.3.56

Linux or Mac

pip
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

Add more Classes(Objects) for Distance Estimation

You will make changes on these particular lines DistanceEstimation.py

python
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

Reading images and getting focal length

You have to make changes on these lines 📝 DistanceEstimation.py
there are two situations, if the object(classes) in the single image then, here you can see my reference image <img src='ReferenceImages/image4.png' width=200>
it has to two object, person and cell phone

python
# 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 👍

python
# 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