Abstract:
Deep Convolutional Neural Network (CNN) has recently made ground-breaking advances on several vision tasks such as objects detection and recognition, classification and semantic segmentation of images. It has achieved state-of-the-art performance on several image recognition benchmarks. The goal of this project is to develop a system capable of detecting and recognizing objects in real time video without substantial memory requirements using Deep CNN. Different deep learning-based methodologies have been proposed to achieve this, and a thorough study of them is undertaken here. A common paradigm to address the problem is to train object detector models with image data sets and apply these detectors in an exhaustive manner across all locations and scales. In this work, saliency-inspired CNN models are used for recognition which predict a set of class-agnostic bounding boxes along with a single score for each box, corresponding to its likelihood of containing object of interest. Python as a programming language, TensorFlow library for computing and OpenCV for computer vision, are used to complete the project. Region-based object detector model such as Faster Convolution Neural Network (Faster-RCNN) inception v2 and MobileNet Single Shot MultiBox Detector (SSD) are used to localize and recognize the objects and compare the accuracy of those predefined models to get the clear concept of model performance in using for different aspects and situations.
