Object detection in live streaming video using deep learning approach

Abstract

Object detection in a complex scene is of challenge as the image contains numerous features. Object detection is important in surveillance, healthcare, recognition, manufacturing, military and agriculture etc. The traditional approaches are limited in accuracy and mainly focussing on image datasets. With the application of deep learning algorithm, the accuracy can be increased. In this paper, the Convolutional Neural Network (CNN) is introduced to detect the objects in the image frames. The input of video is decomposed into image frame and the particular image frame is considered for processing. The image is preprocessed and the noises are removed for further processing. In the next step, feature extraction process contains edge detection and extracting the meaningful shape and texture in the image. The shape which contains the higher level feature is then feed into convolutional layer. The convolutional layer consist of four parts such as number of filter, filter size, pooling layer and max pooling layer. With the application of activation function Relu, the objects can be identified. Furthermore, the Gradient descent algorithm is applied for optimization to improve the accuracy. The system has been tested with ImageNet and the results are very promising.