Convolutional neural network for vehicle detection in low resolution traffic videos

• Main Authors: Bautista, Carlo Migel, Dy, Clifford Austin, Mañalac, Miguel Iñigo, Orbe, Raphael Angelo, Cordel, Macario
• Format: text
• Published: Animo Repository 2016
• Subjects: Vehicle detectors; Neural networks (Computer science); Traffic monitoring; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/3013
• Institution: De La Salle University

Recent works on Convolutional Neural Network (CNN) in object detection and identification show its superior performance over other systems. It is being used on several machine vision tasks such as in face detection, OCR and traffic monitoring. These systems, however, use high resolution images which contain significant pattern information as compared to the typical cameras, such as for traffic monitoring, which are low resolution, thus, suffer low SNR. This work investigates the performance of CNN in detection and classification of vehicles using low quality traffic cameras. Results show an average accuracy equal to 94.72% is achieved by the system. An average of 51.28 ms execution time for a 2GHz CPU and 22.59 ms execution time for NVIDIA Fermi GPU are achieved making the system applicable to be implemented in real-time using 4-input traffic video with 6 fps. © 2016 IEEE.