State-of-the-Art Object Detectors for Vehicle, Pedestrian, and Traffic Sign Detection for Smart Parking Systems

Abstract

To meet the safety requirements of pedestrians and other vehicles in smart parking systems, vehicles rely heavily on visual data to classify and detect target objects. In real-time, deep learning (DL)-based algorithms have shown excellent results in object detection. While several studies have thoroughly investigated various DL-based object detection methods, only a few have focused on multi-detection, encompassing pedestrian detection, vehicle detection, and traffic sign detection. This work provided a comparative analysis of six independent variants of Faster-RCNN and SSD Models for pedestrians, vehicles, and traffic sign detection. There is a lack of comparison leveraging on detailed evaluation metrics among existing models. To address the issue of little or non-availability of datasets for multi-detection, this work provided a mini dataset for task (TraPedesVeh). Experimental results analyzed using detection accuracy, average precision (AP), average recall (AR), training time, and loss show that the Faster-RCNN model with the Inception backbone outperformed all other models.