Performance Analysis of Object Detection Algorithm for Intelligent Traffic Surveillance System

Object detection algorithm such as convolution neural networks (CNN) is implemented for traffic surveillance applications. A neural network consists of input minimum one hidden and an output layer. Urban vehicle dataset, which consists of four classes of images such as Heavy, Auto, Light, and Two-wheeler captured during day, evening and night, which includes blur images. The dataset is composed of images of varying illumination. Performance parameters such as accuracy, precession, recall and f1 score calculated for night and blur image dataset. Obtained results shows that the algorithm effectively detects objects with an accuracy of 91% for night images and 88 % for blur images.