ABSTRACT
Vision-based vehicle detection is the backbone of intelligent transportation systems. Accelerated research in a convolutional neural network (CNN) has achieved improved results related AU: Is edit okay here? to the computer vision-based conventional methods. Overgrowing demand for real-time object detection in the intelligent transportation systems gave rise to the numerous deep learning techniques, enabled to recognize the objects with optimum training cycle. Recent algorithms based on region proposal methods influence object detection by sharing the convolutional features of the complete image with the detection framework. Region proposals network generates region proposals which are fed to the region of interest (RoI) pooling layer, giving the end-to-end object detection. Experiments are performed on different models of faster R-CNN. Faster R-CNN and single-shot detectors (SSD) architecture are validated on Miovision traffic camera (MIO-TCD) database. The improved faster R-CNN model performs better than the SSD models when evaluated based on object detection rate and objectness score.
