Object Detection And Training Of Deep Neural Networks

Abstract

Humans always want their daily basis tasks to be done without any intervention. Everything around us is filled with tremendous amount of perceivable information. As the technology advances, the amount of knowledge that we can obtain from that information also increases. And so, computer vision and artificial intelligence were introduced. Computer vision and artificial intelligence are one among the busy fields of technology in which advancements are constantly being introduced. Computer vision is a branch of science of computer systems which can recognize as well as understand images. Detecting and recognizing objects in unstructured as well as structured environments is one of the most challenging tasks in computer vision and artificial intelligence research and is one of the aspects of computer vision. We can use the science of object recognition for many useful applications in order to enhance the knowledge gained from the visible information around us.

This paper presents a trainable architecture of neural networks that can detect as well as recognize any object by using appropriate object detection algorithms and a lenses or web camera. Mobilenetv2 is a neural network architecture that uses depth wise separable convolution as effective building blocks to scan, classify and detect objects from an image or a scene. Mobilenetv2 is used a extractor of classified elements based on their respective feature and provides an efficient mobile oriented model to be used as a base for many image recognition tasks.