Search

Road accidents are major cause of death which has been increased by 46% since 1990. In recent years, significant efforts have been invested in four-wheeled vehicle detection that improved intelligent transportation systems and decreased the calamity rate. However, the automatic detection of two-wheeled vehicles remains challenging due to occlusion, illumination variation, environmental conditions, and viewpoint variations. In this paper, we present a comprehensive methodology of two-wheeled vehicle detection using two categories of deep learning-based object detection models including two-step and single-step techniques. In two-step object detection techniques, experiments are carried out with object detection models such as a region-based convolutional neural network (RCNN), Fast-RCNN, Faster-RCNN, and region-based fully convolutional network (R-FCN), while in single-step object detection techniques, detection is performed using the single-shot multibox detector (SSD), SDDLite, and you only look once (YOLOv3) detection models. The performance of the proposed methodology is evaluated on two benchmark datasets, i.e., MB7500 and Tsinghua-Daimler Cyclist data. The experimentation results demonstrate that Faster-RCNN with the Inception-Resnetv2 backbone model impressively outperforms two-step object detection techniques, while in single-step object detection techniques, SSD with the Inceptionv2 model shows superior performance. Further, the performance comparison of the proposed methodology with existing state-of-the-art methods confirms its effectiveness in two-wheeled vehicle detection. [ABSTRACT FROM AUTHOR]