OpenCV Pedestrian Detection Application Solution Based on MYIR's T527 Development Board

In this article, we will introduce the testing of OpenCV pedestrian detection solution based on MYIR's MYD-LT527 Development Board (based on Allwinner T527 processor).

MYD-LT527 Development Board Top-view

MYD-LT527 Development Board Bottom-view

1. Software Environment Installation

1.1 Install OpenCV

sudo apt-get install libopencv-dev python3-opencv

1.2 Install pip

sudo apt-get install python3-pip

2. Introduction to Pedestrian Detection

The key steps in building the pedestrian detector using HOG and SVM include:

• Prepare Training Dataset: the training dataset should contain large number of positive samples (pedestrian images) and negative samples (non-pedestrian images).

• Compute HOG Feature: for each image, compute HOG feature. HOG feature is a one-dimensional vector where each element represents the gradient intensity at a specific location and direction in the image.

• Train SVM Classifier: using the HOG feature as input, train the SVM classifier which will learn to distinguish between pedestrians and non-pedestrians.

• Evaluate the Model: evaluate the trained model using the test dataset. Calculate metrics such as accuracy, recall and F1 score of the model.

3. Code Implementation

import cv2

import time

def detect(image,scale):

imagex=image.copy( ) # Make a copy of the function internally so that each function runs on a different image;

hog = cv2.HOGDescriptor( ) # Initialization direction gradient histogram description;

#Set up the SVM as a pre-trained pedestrian detector;

hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector( ));

# Invoke the function detectMultiScale to detect the corresponding border of the pedestrian;

time_start=time.time( );# Record start time;

# Access (rectangular boxes corresponding to pedestrians, corresponding weights);

(rects, weights) = hog.detectMultiScale(imagex,scale=scale);

time_end = time.time( ) # Record end time;

# Draw each rectangular box;

for(x, y, w, h) in rects:     

cv2.rectangle(imagex, (x, y), (x+w, y+h), (0, 0, 255), 2)

print("sacle size:", scale, ",time:", time_end-time_start)

name=str(scale)

cv2.imshow(name; imagex) # Show original effect

image = cv2.imread("back.jpg")

detect(image,1.01)

detect(image,1.05)

detect(image,1.3)

cv2.waitKey(0)

cv2.destroyAllWindows( )

4. Practical Operations