IoT intelligent video logarithmic imaging and node analysis technology analysis

Video analysis applications in the Internet of Things (IoT) can be improved using node analysis and logarithmic imagers. Video analytics applications attempt to take advantage of the rich information resources in the everyday world for a number of reasons, including face recognition for daily monitoring, but most of the reasons focus on predictive analytics and behavioral analysis. The information gathered in these applications can be processed at a higher level through cloud computing. However, depth processing has its limitations and can be improved in many ways by adding node analysis and logarithmic imagers to the combination.

Data analysis can be improved by adding node analysis to the combination to mitigate communication with the cloud. The bandwidth requirement for cloud computing is two orders of magnitude (if not three) more than the node analysis application. Therefore, node analysis requires less computing power and reduces latency. Densely populated markets, traffic-stricken areas, and urban parking lots are some of the most complex environments that can be detected using node analysis for predictive and behavioral analysis. Advanced processing of these environments in the cloud helps to develop business strategies, channel traffic, and improve the efficiency of government-managed parking lots. However, using low-end software at the sensor nodes instead of performing cloud analysis can improve the latency, bandwidth, security, and power consumption of these scenarios.

In addition to intelligentization at the nodes, adding logarithmic imagers to the combination also has advantages in areas where conventional imagers do not meet the requirements, thereby enhancing system functionality. In addition to reducing the dependence on brightness variations, the log-imager provides a higher image processing dynamic range. For example, in shadows, reflections, sudden changes in light, and high-contrast scenes, log-imagers perform better than traditional imagers. In video applications, the resolution of these problems facilitates data capture, thereby enhancing the analytical capabilities of the nodes. By improving data capture capabilities, the entire video analytics application can be significantly improved.

Improvements in node analysis techniques and log-imager implementations help solve video analytics applications in the Internet of Things. Security, decision making latency, data bandwidth, and computing power are some of the engineering challenges that are common in IoT applications. These engineering problems can be greatly reduced by reducing data transfers, which is why node analysis is attractive for IoT applications. In video analytics applications, limited contrast and brightness dependence are challenges that need to be addressed together. The logarithmic imager is the key to video analytics applications and can solve this problem almost. In general, the use of node analysis techniques and log-imagers enhances video analytics applications in the Internet of Things.

By processing data based on expected visual events, measurement data can be quickly converted to appropriate actions without having to transmit any data to the cloud server or transfer small amounts of data. By quickly analyzing video data instead of passing it to the cloud, decisions can be made locally to improve system latency. By reducing the transmission of data with the risk of interception, not only can the decision delay be significantly reduced, but security can also be improved.

Only the most valuable information needs to be transmitted to the cloud outside the node for prediction or behavioral processing. Optimized data partitioning leverages cloud value because video analysis frames with full bandwidth are typically not required. On a fixedly mounted camera, most of the visual data between frames is static data and can be filtered at the node. Edge node video analysis provides a variety of filtering interpretations to distinguish between various expected object types: cars, trucks, bicycles, humans, and animals. This decimation operation reduces the data bandwidth and associated computing power required on the cloud server, and if the full frame rate video data sent downstream is analyzed, it will consume a large amount of data bandwidth and computing power. This reduction in bandwidth can achieve two or three orders of magnitude improvement over cloud computing applications, a key performance improvement in node analysis technology implementation.

Video analytics applications can be further improved by replacing log imaging techniques with common problems associated with traditional imagers. Most conventional imagers are linear imagers that use pixel-generated voltage as a linear function of light, and such pixels result in limited contrast. The linear imager also utilizes a uniform exposure phase that limits its dynamic range relative to pixel exposure time over the frame rate range. Finally, the contrast of conventional imagers depends on the brightness, which can lead to reflection-related contrast problems. These common problems can be eliminated by replacing the logarithmic imager with the voltage generated by the pixel as a logarithmic function of the light. See Figures 1, 2 and 3 for the difference between a conventional imager and a logarithmic imager.

Figure 2. ADIS17002 logarithmic image exposure

Figure 3. ADIS17002 logarithmic image exposure using built-in edge detection

Some conventional imagers are trying to solve the contrast-related problems that prevent users from fully capturing their target environment data. These contrast problems stem from the linear nature of the generated voltage in each pixel. The voltage generated in a linear imaging pixel is proportional to the number of photons illuminated; therefore, its dynamic range is limited compared to logarithmic imaging. Reducing the contrast associated with these linear imagers means reducing the dynamic range. Contrast reduction can adversely affect the analysis of sensor nodes in IoT applications, ultimately affecting the overall performance of the system. The logarithmic imager provides a wider range of brightness levels, thereby increasing the contrast produced by the logarithmically generated pixel voltage. However, an increase in contrast results in higher photosensitivity, which may not be a desirable effect in some applications. Alternatively, an increase in sensitivity may also be an advantage, depending on the application.

Reflections in sunny or bright environments may further hinder video capture using traditional imagers. For example, when there is reflection on the windshield, face recognition inside the car becomes more difficult. This video capture barrier can introduce errors into the system or lose important data, which can adversely affect video analytics. Since the contrast between the pixels of the linear imager depends on the brightness, it is reflective; therefore, the reflection of the linear imager is more prominent. This dependence on brightness is given in Equation 1.

In addition, the contrast of the log-imager is independent of brightness due to its own logarithmic properties, helping to reduce reflections or sudden changes in light. See Figure 2 for the characteristics of the logarithmic imager that are independent of brightness.

To provide platform-level solutions, Analog Devices is moving beyond the development of individual components; these solutions help customers quickly deploy proven, intelligent solutions that deliver higher performance at lower system cost. Smart applications start with accurate and reliable data, which is available through ADI's advanced detection and measurement capabilities. In addition, ADI works with customers to develop unique system-level solutions that solve all problems. The ADIS1700x is one of the solutions that enables quarter of video graphics array (QVGA) imaging analysis.

Figure 4. Functional block diagram

The ADIS1700x is a small, logarithmic sensitive QVGA analysis imager module with digital signal processing that optimizes video performance. In addition to accelerometers for image stabilization, tilt and shock detection, the module uses a low-power Blackfin processor for node analysis. In addition, it uses built-in edge detection technology to track and calculate object motion. Unlike traditional imagers, the logarithmic imager has a unique exposure phase for each 14 μm × 14 μm pixel. The conformal coating for outdoor operation makes it ideal for large-scale deployments, creating emerging smart city and building applications. The ADIS17001 is equipped with a 110° field of view (FOV) lens, while the ADIS17002 is equipped with a 67° FOV lens. These two options are suitable for a variety of target applications, including parking lot monitoring, parking violation enforcement, traffic length detection, and industrial analysis.

Figure 5. ADIS17002: diagonal direction (left), board lens side view (middle) and back side (right).

In general, the use of node analysis techniques and logarithmic imagers can significantly improve video applications in the IoT space, which is the approach that Analog Devices has introduced with the ADIS1700x module. Node analysis, not cloud computing, is conducive to the development of IoT applications. Logarithmic imagers have the advantage of being unmatched by their competitors, further improving IoT applications. In summary, the video analysis application in the Internet of Things field combines with node analysis technology and logarithmic imager to form a robust system-level solution.