Intelligent Vision with TensorFlow using Neural Network Algorithms

Abstract

Computer vision and video analytics are the torrid research area in Machine learning and their establishment process traditionally starts with object detection and eventually tracking. In recent years, there is a tremendous growth in performing comprehensive study based on the field of object detection and Pattern Analysis. In our system we have improvised and experimented with detection method based on machine learning and deep learning approach in object recognition and pattern analysis. We assume object detection as a retrogression problem to spatially separated corresponding class probabilities and bounding boxes. Many prominent algorithms have been designed for object detection, Pattern Analysis and tracking, which also includes edge tracking, color segmentation and pattern matching. A single neural network is capable of predicting class probabilities and bounding boxes directly from the full image per cycle. Therefore we have used various neural network algorithms such as YOLOv3, Single Shot Multiple detection algorithm to carry out video analysis using object detection and drowsiness detection using pattern or behavior analysis with the help of Tensorflow. The framework will recognize object continuously, from the input perceived through camera where it can apparently capture a required frames to predict the object and also to match the pattern. It has been accomplished using real-time video processing and a single camera. The proposed system is versatile to operate in complex, real time, non-plain environment.