Improved pedestrian detection algorithm based on HOG and SVM

Pedestrian detection becomes an acknowledged challenging problem to the development of intelligent video surveillance and vehicle active safety. At present, there are some shortcomings in pedestrian detection, such as missing detection, false detection, inaccurate detection frame location. An improved algorithm based on Histograms of Oriented Gradients (HOG) and Support Vector Machine (SVM) is proposed to tackle these problems. This algorithm employs face detection technology to further enhance the accuracy of pedestrian detection. Firstly, face detection and pedestrian detection are adopted to accurately locate the face contour and the human body contour, respectively. And then, detection frame is redraw by synthesizing the coordinates of the upper left corner of the face contour and the coordinates of the lower right corner of the human body contour. A new test database is established with numerous images in complex scenes. And then, relevant experimental results demonstrate the effectiveness of proposed algorithm and shows better target detection accuracy and real-time performance compared with existing methods. © 2020 Computer Society of the Republic of China. All rights reserved.