Welcome to the new course on "Real-time Human Detection using Tensorflow". A novel approach has been proposed to achieve the human detection in photos, videos, along with real-time detection using the system webcam and via the external camera.
Surveillance is an important aspect in security, especially with places like banks, jewellery shops and so on. Now-a-days, even small hotels and bakeries use surveillance systems to protect their properties. But, counting people in visual surveillance is hard and challenging problem. Automatic counting surveillance of individuals in public areas is vital for safety control. Previously many techniques and methods were proposed but it was not very successful.
There is an ever-increasing amount of image data in the world, and the rate of growth itself is increasing. Going beyond consumer devices, there are cameras all over the world that capture images for automation purposes. Cars monitor the road, and traffic cameras monitor the same cars. Robots need to understand a visual scene in order to smartly build devices and sort waste. Imaging devices are used by engineers, doctors and space explorers alike. To effectively manage all this data, we need to have some idea about its contents. Automated processing of image contents is useful for a wide variety of image-related tasks. And, this helps the developers and computer programmers to come with more optimized solutions.
For computer systems, this means crossing the so-called semantic gap between the pixel level information stored in the image files and the human understanding of the same images. Computer vision attempts to bridge this gap. Objects contained in image files can be located and identified automatically. This is called object detection. As we will demonstrate, convolutional neural networks which are currently the state-of-the-art solution for such type of object detection. The main task of this project is to review and test convolutional object detection methods. In the theoretical part, we review the relevant literature and study how convolutional object detection methods have improved in the past few years. In the experimental part, we study how easily a convolutional object detection system can be implemented in practice and testing.
Every problem needs an AI/ML solution. The excitement has led to high expectations. Hence, in this python project, we are going to build the Human Detection and Counting System through Webcam. This is actually an intermediate level deep learning project on computer vision and tensorflow, which can assist you to master the concepts of AI and you will be an expert in the field of Data Science.
So, for your easy understanding, the course has been divided into 14 sections. Then, let us see what we are going to learn in each section.
In the first section, we will learn about Artificial Intelligence, Neural Networks, Object Detection Models, Computer Vision Library, TensorFlow, TF API, and its detailed specifications and applications along with appropriate examples.
In the second section, we will learn about Human Detection Model and then we’ll understand how to install software and tools like, Anaconda, Visual Studio, Jupyter, and so on. Next, we will learn about the IDE and the required settings. Later, this will help us to understand, how to set up python environments and so on.
Testing small programs separately in jupyter notebook will give you clarity about the functionality and the working principle of jupyter notebook. So, in the third section, we will learn about setting up jupyter notebook and workspace.
The fourth section begins with importing dependencies, defining and setting paths for labels, and real-time demonstrations along with source code.
In the fifth section, we will get to know about computer vision library and how to capture images using opencv. We will understand the script step by step and then proceed further with real-time demonstration and image labelling tool. Thereafter, we will learn about Annotations and its types. And finally, we’ll start making annotations.
In the sixth section, we will start with the Human Detection Model. Then, we’ll learn to customize our own model. Thereafter, we will proceed with pretrained models, script records, label maps, and so on. After that, we’ll start working with workspace.
Next section will teach us about TensorFlow Model API and Protocol Buffers. Here, we’ll proceed with Model Garden, WGET Module, Protoc and the verification of the source code. Then we’ll learn here, how to download pretrained model from TensorFlow Zoo.
After that, in the 8th section, We’ll work with models. Here, we’ll learn how to create label map, how to write files, and so on. Then, we’ll learn about model records like training and test records, copying model config into training folder along with real-time demonstration.
In this 9th section, we’ll proceed with pipeline configurations, where we’ll learn about checkpoints. Next, we’ll go ahead with configuring, copying and writing pipeline config. And at last, we’ll do the verifications.
In the 10th section, you will understand how to train and evaluate Human Detection Model. Here we’ll proceed with Training Script, commands for training, and verifications. This is the most important section where we’ll build our Human Detection Model. And, we’ll have to be very careful at this stage. Because, Training may take a long hours or a day, if your system doesn’t have any GPU and have used a higher training steps. After completion of trainings, model evaluation step comes. So here, we’ll understand about model evaluation, mean average precisions, recalls, confusion matrix and so on.
11th section will take you to the trained model and checkpoints. Here, we’ll learn about loading pipeline configs, restoring checkpoints and building a detection model. And then, we’ll understand the source code.
In the 12th section, we will get to know, how to test Human Detection Model from an image file. Here, we’ll import recommended libraries, and then learn about category index, defining test image paths and so on.
13th section will get your hands dirty. You will do real-time detections from a webcam and will get to know, how the model performs.
Finally, in the 14th section, we’ll understand about freezing graphs, tensorflow lite and archive models. This is the last section, where we’ll save our Human Detection Model by using freezing graph method. Then we’ll learn how to convert Human Detection Model into TensorFlow Lite model. And then, we’ll end this project by archiving our model for future editing.
