Explore video and image analysis, monitoring, and security in cloud expansion

The author describes how to use OpenCV and similar tools for digital video analysis and methods to extend such analysis using cluster and distributed system design. In previous installments, a coprocessor designed specifically for video analysis and new OPENVX hardware acceleration was discussed, which can be applied to the computer Vision (CV) sample provided in this article. This new data-centric CV and video analysis technology requires system designers to rethink application software and system design to meet demanding requirements such as real-time monitoring and security for large, public facilities and infrastructure, and a more entertaining, interactive, and secure world.

The use of standards (such as those from the Motion picture experts Group, or MPEG) to encode video for compressing, transmitting, decompressing, and displaying digital video has led to dramatic changes in the computing industry, From social networking media and amateur digital theaters to improved training and educational change. Tools for decoding and using digital video are widely used every day, but video analysis requires tools for encoding and analyzing uncompressed video frames, such as Open Computer Vision (OpenCV). An easily accessible and powerful digital Video codec tool is FFmpeg; for static images, GNU image 處理 (GIMP) is useful. With these 3 basic tools, open source developers are fully able to start exploring computer vision (CV) and video analysis. However, before you analyze these tools and development methods, let's start by providing a better definition of these words and think about what they are used for.

In the first article in this series, cloud extensions, part 1th: Building your own cloud and using on-demand HPC extensions provides a simple example of using OpenCV, which enables Canny edge conversion on continuous real-time video from the Linux®web camera. This is an example of a CV application that you can use as the first step in segmenting an image. In general, CV applications involve the acquisition of digital image formats, images and image sequences (films), processing and conversion, segmentation, recognition, and final scene descriptions, which represent the image elements of the illuminance point. The best way to understand the purpose of the CV is to view the sample. Figure 1 shows a facial feature detection analysis using OpenCV. Note that in this simple example, the Haar Cascade method (a machine learning algorithm) is used to perform instrumentation analysis. The algorithm can most accurately detect the face and eyes that have not been blocked (for example, when my youngest son's face is turned to one side) or that has been obscured by shadows, and the face and eyes when the subject is not squinting. The following may be the most important comment on the CV: it is not a simple question. Researchers in this field have often noticed that although it has made great strides since its birth more than 50 years ago, most applications still fail to catch up with a 2-year-old's ability to distinguish and recognize the scene, especially when it comes to the ability to generalize and perform recognition under a wide range of conditions (illumination, size change, direction and environment).

Figure 1. Using OpenCV to perform facial recognition

To help you understand the profiling methods used in the CV, I created a smaller set of tests for the Alaska State Anchorage area image, which can be downloaded. These images have been processed using GIMP and OpenCV. I developed the C + + code to use the OpenCV application programming interface for Linux WEB cameras, pre-captured images, or MPEG videos. Using CV to understand video content (image sequences, whether in real time or from a pre-collected image sequence database) is often called video analysis.

Define video Analysis

Video analysis is broadly defined as the analysis of digital video content from a camera (usually visible, but may also come from other parts of the spectrum, such as infrared) or stored image sequences. Video analysis involves a number of disciplines, including at least:

Image acquisition and coding. The image is collected and encoded in a series of images or compressed image groups. This phase of video analysis can be complex, including photometer (camera) technology, analog coding, the number format of light sampling (pixel) arrays in frames and sequences, and methods to compress and decompress this data. CV. The inverse of the image rendering, where the captured scene is converted to a description, in contrast to a description being rendered as a scene. CV assumes that the process of using a computer for "viewing" should be performed anywhere humans can view, which is usually distinguished from machine vision. Viewing like a person often means that the CV solution uses machine learning. Machine vision. The same is the inverse of the rendering process, but it is most often used for process control purposes in tightly controlled environments, such as checking printed circuit boards or finished parts to ensure that they are in a geometric error-tolerant range. Image processing. The digital signal processing method can be widely applied to the photometric and radiometer (detectors for measuring electromagnetic radiation) in order to understand the properties of the observed objects. Machine learning. Through training data, an algorithm is improved to develop some new algorithms, which can improve the performance and popularize the application when using new data. Real-time and interactive systems. These systems need to respond to a service request within a deadline, or at least meet the quality of service that is in line with the SLA that is signed between the service customer or the user. Storage, networks, databases, and computations. These are all needed to deal with the digital data used in video analysis, but a subtle and important difference is that this is an intrinsic, data-centric computing problem, as described in part 2nd of this series.

As a result, video analytics is wider in scope than CV and is a system design problem that may include mobile elements such as smartphones (Google Goggles) and cloud-based services for the overall system CV. For example, IBM has developed a video analytics system called the VCAs (video correlation and analysis suite, visual Association and Analytics Suite), IBM Travel and Transportation Solutions Newsletter Smarter Safety and security Solution for Rail [PDF]; it's a good example of a system design concept. A detailed description of each system design discipline involved in a video analysis solution is outside the scope of this article, although resources provide clues to more information for system designers. The remainder of this article focuses on the CV processing examples and applications.