Performance evaluation of CNN-based pedestrian detectors for autonomous vehicles

Abstract

With the widespread application of deep learning methodologies, many fields including Intelligent Transportation Systems (ITS) have integrated neural network-based models. In return, the promising performance of neural network-based models attracts more research efforts being paid to the deep learning area. Autonomous vehicles, as the future ITS participants, have achieved tremendous development with the help of convolutional neural network (CNN)-based detectors (e.g., vehicle detectors, lane detectors, pedestrian detectors, traffic sign detectors, etc.) in recent years. However, we have noticed that researchers leveraged different computing power when publishing their experimental results, which could lead to unfair comparisons. In this paper, we focus on CNN-based pedestrian detectors. We conduct a comprehensive comparative study of the representative CNN-based pedestrian detectors, aiming to investigate the influence of experimental settings by eliminating the bias of different experimental environments.