Chapter 15: An Effective Pedestrian Detection Using Convolutional Neural Network: Computer Vision Applications

Automated techniques must be used to handle the vast amounts of data provided by today’s big number of surveillance cameras since they are unable to analyze the information manually. For pedestrian detection, the vast majority of presently known techniques cannot handle this volume of data in real time. A new strategy based on the maximum search problem theorem is proposed in this paper to improve pedestrian detection algorithms by randomly selecting a limited number of detection windows from among all available detection windows. Although random filtering may choose areas that catch every individual in a picture, certain windows can only cover sections of a person, reducing the accuracy. Using a regression model, the windows may be resized based on their position. The suggested technique does not need any processing when picking windows. Partial least squares-based pedestrian identification is successful in accuracy and less computational cost, according to the studies.