Object Detection System Based on Convolution Neural Networks Using Single Shot Multi-Box Detector

Abstract

Abstract In this paper we propose object detection technique to detect objects in real time on any device running the model and in any environment. Object detection and training is a vast, vibrant and yet inconclusive and complex area of the computer vision. In this proposed work, convolutional neural network are used to develop a model which is composed of multiple layers to classify the given objects into any of the defined classes. The proposed schemes then use multiple images to detect the objects and label them with their respective class label. These objects are detected by making use of higher resolution feature maps. This is possible because of the recent advancement in deep learning with image processing. These images can be from the video frames which are fed into the model. Our scheme uses separate filters with different default boxes to tackle the difference in aspect ratio and also used multi-scale feature maps for object detection. The training of the model takes place until the error rate is less. The trained model is used to test some sample images. To speed up the computational performance of object detection technique we have use single shot multi-box detector algorithm along with the help of architecture of faster region convolutional neural network. The accuracy in detecting the objects is checked by the different parameters like loss function (LP), mean average precision (mAP), frames per second (FPS).