Vision-based vehicle detection and counting system using deep learning in highway scenes

Abstract

Intelligent vehicle detection and counting are becoming increasingly important in the field of highway management. However, due to the different sizes of vehicles, their detection remains a challenge that directly affects the accuracy of vehicle counts. To address this issue, this paper proposes a vision-based vehicle detection and counting system. A new high definition highway vehicle dataset with a total of 57,290 annotated instances in 11,129 images is published in this study. Compared with the existing public datasets, the proposed dataset contains annotated tiny objects in the image, which provides the complete data foundation for vehicle detection based on deep learning. In the proposed vehicle detection and counting system, the highway road surface in the image is first extracted and divided into a remote area and a proximal area by a newly proposed segmentation method; the method is crucial for improving vehicle detection. Then, the above two areas are placed into the YOLOv3 network to detect the type and location of the vehicle. Finally, the vehicle trajectories are obtained by the ORB algorithm, which can be used to judge the driving direction of the vehicle and obtain the number of different vehicles. Several highway surveillance videos based on different scenes are used to verify the proposed methods. The experimental results verify that using the proposed segmentation method can provide higher detection accuracy, especially for the detection of small vehicle objects. Moreover, the novel strategy described in this article performs notably well in judging driving direction and counting vehicles. This paper has general practical significance for the management and control of highway scenes.

A Novel Deep Learning Based Method for Detection and Counting of Vehicles in Urban Traffic Surveillance Systems

“…The authors report an average counting accuracy of 90% in three different test scenarios; however with the increase of vehicles, the tracking accuracy begins to decrease. In (Song et al, 2019) a method is proposed for the problem in detecting and counting small vehicles, this method foregrounds the surface of the highway dividing it into a remote area and a proximal area. These areas are the inputs of the YOLOv3 model for the detection of the car, bus and truck categories.…”

“…Although hardware solutions have a higher counting precision than software solutions, these sensors have limitations to obtain detailed information on the behavior of the vehicular flow, in addition to being intrusive and presenting high costs of installation and maintenance. On the other hand, software-based systems, especially video-based methods that perform image processing (computer vision) have started to stand out because it is an inexpensive, non-intrusive approach that have proven to be successful (Song et al, 2019).…”

Enhancing City Sustainability through Smart Technologies: A Framework for Automatic Pre-Emptive Action to Promote Safety and Security Using Lighting and ICT-Based Surveillance

“…The success of the ImageNet project has prompted many endeavors associated with self-driving technology using CV-real-time traffic light recognition [38], pedestrian detection [39]-just to name a few. The use of CV in a smart city context [40] finds its usefulness in vehicle recognition in traffic scenes [41], which could be extended to microscopic behavior analysis of pedestrians and cycle commuters in shared spaces [42]. CV is also deployed in a drone's vision [43], which when paired with GIS [44] can present endless opportunities.…”

A Real-Time Vehicle Detection System under Various Bad Weather Conditions Based on a Deep Learning Model without Retraining

“…Numerous vehicle detection methods have been proposed to obtain trustworthy traffic data for the development of intelligent traffic systems. Briefly, we can split these methods into two categories: the first category adopts the classical machine learning models trained from manually defined features [ 1 , 2 , 3 , 4 , 5 , 6 ], such as Haar-like features and histogram of gradients (HoG) features with AdaBoost classifiers and support vector machines (SVM) classifiers; the other category comprises deep learning model-based approaches constructed from large amounts of labeling data [ 7 , 8 , 9 , 10 , 11 ]. Both categories can currently satisfy real-time requirements, yet the detection and classification accuracies of deep learning model-based approaches are still more robust than machine learning models constructed from manually defined features.…”