Improved Center and Scale Prediction-Based Pedestrian Detection Using Convolutional Block

Donggeun Kim, Sangwoo Park, Donggoo Kang, Joonki Paik

Abstract: Pedestrian detection is an important computer vision algorithm in advanced driver assistance systems (ADAS). In this paper, we propose a deep learning-based pedestrian detection method that is specialized for ADAS. The proposed model improves both accuracy and processing speed by applying a convolutional block attention module (CBAM) to center and scale prediction (CSP)-based detector. To evaluate the detection performance, we tested the proposed method using Caltech dataset, which is one of the most commonly used datasets for pedestrian detection challenges. Our model achieves 7% miss rate and takes 40.5 ms to inference an image.