Performance Evaluation of Pedestrian Detectors for Autonomous Vehicles

With the development of deep learning methods, adopting CNN-based detectors has become a trend to handle the detection task. The proposal of Intelligent Transportation Systems (ITS) has once again brought autonomous vehicles into the public eye. Pedestrian detection can ensure pedestrians' safety, and it is considered one of the most challenging problems that urgently need to be solved. We have noticed that researchers use various environments when publishing experimental results, leading to unfair comparisons of experimental results. Under different computing resources, the performance of the detector may be weakened or enhanced. In this paper, we will compare two representative detectors with the same computing power for a fair comparison study, aiming to find out how experimental settings affect the detector's accuracy.