Crowd Counting with Deep Learning (Mini Project)

Pretrained CNN model

• Model

  Methods followed

  • Non Maximum Suppression : is a key step in many computer vision applications. It is a class of algorithms to select one entity (e.g., bounding boxes) out of many overlapping entities. The most common approach for NMS for object detection is a greedy, locally optimal strategy with several hand-designed components (e.g., thresholds).

  • Confidence/Consistency Map : It basically is a probability density method to know how much an image is similar to the image prior to it. It assigns each pixel of the new image a probability, which is the probability of the pixel color occurring in the object in the previous image. More algorithms like this are ensemble tracking, CAMshifts, Kalman filter, mean-shift.

  • Gamma Correction : This controls the overall brightness of an image. Gamma values less than 1 will shift the image towards the darker end of the spectrum while gamma values greater than 1 will make the image appear lighter.

  • Downscaling : Downscaling is any procedure to deduce high-resolution information from low-resolution images. This is used in case of images with low resolution.

  • CNN (Convolutional Neural Network) : This is a type of Neural network whose use is generally to analyze imagery. It uses a special technique called Convolution meaning it transforms an image using every pixel and their local neighbours.

  • ReLU (Rectified Linear Unit) : It is an activation function which will output the input directly if it is positive, otherwise, it will output zero (Absolute classification). It is more effecient and time saving than Sigmoid function.

Mounting

• Mount GDrive account to colab

Imports

import numpy as np
import matplotlib.pyplot as plt
import pretrainedmodel as cnn
import cv2
import torch

Images

• Source

Importing the Data and Image

data = '/content/drive/MyDrive/weights/weights.pth'

image = cv2.imread('/content/crowd2.jpg')

Detecting Heads

• To detect the heads and count them, use .head_detection() function.

Counting

• Use .sum() function to count the heads in the image.