CONVOLUTIONAL NEURAL NETWORK STRATEGIES FOR REAL-TIME OBJECT DETECTION

CONVOLUTIONAL NEURAL NETWORK STRATEGIES FOR
REAL-TIME OBJECT DETECTION

Prof. Dr. Sonali Kadam1, Aastha Sharma2, Anjali Bari3, Bhumika Trivedi4, Amruta Shinde5
1Professor, Department of Computer Engineering, Bharati Vidyapeeth’s College of Engineering for Women, Pune
2,3,4,5Student, Department of Computer Engineering, Bharati Vidyapeeth’s College of Engineering for Women, Pune
---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract - Segmentation, extraction of features, and object recognition from picture data are all areas where computer vision excels. Object detection is attracting a lot of attention from a variety of industries, including healthcare, traffic monitoring, surveillance, robotics, and so on. Because of its employment in sensitive fields, the capacity to identify the item more accurately is a key element. Recent improvements of sdeep learning approaches provide appealing performance to fine-grained picture classification which tries to identify ssubordinate-level categories. This job is exceedingly tough because of strong intra-class and minimal inter-class variation. Convolutional Neural Networks have recently endeavored to deliver a greater degree of efficiency and accuracy in all of the sectors where they have been used, the most prominent of which being Object Detection, Digit, and Image Recognition. It uses a well-defined set of procedures to follow, including Backpropagation, Convolutional Layers, Feature Formation, and Pooling. In addition, this post will look at how to leverage several frameworks and tools that employ the CNN model.
Key Words: Convolutional Neural Network, Fast R-CNN, Activation map, YOLO, Feature detector, Single Shot Detector.