Deep Learning in Vehicle Detection Using ResUNet-a Architecture

Abstract

Vehicle detection is still a challenge in object detection. Although there are many related research achievements, there is still a room for improvement. In this context, this paper presents a method that utilizes the ResUNet-a architecture – that is characterized by its high accuracy - to extract features for improved vehicle detection performance. Edge detection is used on these features to reduce the number of calculations. The removal of shadows by combining color and contour features - for increased detection accuracy - is one of the advantages of the proposed method and it is a critical step in improving vehicle detection. The obtained results show that the proposed method can detect vehicles with an accuracy of 92.3%. This - in addition to the obtained F-measure and η values of 0.9264 and 0.8854, respectively - clearly state that the proposed method - which is based on deep learning and edge detection - creates a reasonable balance between speed and accuracy.