Automatic Vehicle Recognition and Multi-object Feature Extraction Based on Machine Learning

Abstract

Over the past few years the influence of surveillance has left a massive impact on the social lives of people. The introduction of controlling the surroundings with the surveillance system would provide zero error and faster reaction time. In this proposed work, we have developed a system that will detect and analyze the traffic signal images. Thousands of traffic signal images are fed into the computer and trained based on the margins of particular classes. A weight file is generated from this training process. YOLO (you only look once) is an algorithm used for training the images and detecting the images. It is a network for object detection. In this project, the objects are vehicles and human beings. The identification of objects is done by searching the location on the image and arrange those objects with its prediction level. Existing methods like R-CNN and its variations, used a pipeline methodolgy to analyze and segment the images in multiple steps. Accuracy and speed of recognition is very slow in existing methodologies because of the individually done component training. Proposed methodology with YOLO is performed by a unique neural network. The output describes the computation and name of the classes from a fed image, which is used as vehicles and human beings.