PEDESTRIAN DETECTION AND CLASSIFICATION USING DEEP LEARNING

Abstract

In this study, the main contribution is to solve the task of pedestrian detection and adult / kid classification by using two approaches. In the first one, the task is divided into two sub-tasks: pedestrian detection and adult / kid classification. Pedestrian image regions are cropped from input images and passed through a classifier to determine if they are adult images or kid images. Specifically, the pedestrian detection task is studied by using an object detection model YOLO while the classification task is studied by using typical deep models: VGG, Inception, ResNet and EfficientNet. In the second approach, only one object detection model, YOLO is used to detect and classify pedestrians. The obtained results are quite good for both approaches. The first one has a good mean average precision of the pedestrian detection task at 0.797 and the classification accuracy is 0.955. However, the second approach has much better results with a higher mean average precision 0.851 and a much better performing time compared to the first approach.