All-target detection via intelligence on the edge
Source: asmag.com.cn
Date: 2019/12/30
Zeno-Tech ZN-SNC-IB6200WS-I5ZPF (AO)-GLST integrated bullet network smart camera
Compared with cloud-side intelligence, edge-side intelligence has advantages in privacy protection, broadband transmission, response speed and power consumption. At present, artificial intelligence at the edge is becoming the standard for surveillance cameras. Surveillance cameras with multiple deep learning algorithms can achieve full-target detection, tracking and capture; compared with single-dimensional detection cameras, they are more suitable for complex application requirements in smaller projects. This time, a&s tests Zeno-Tech’s ZN-SNC-IB6200WS-I5ZPF (AO)-GLST network intelligent camera, which comes with built-in deep learning-based face, human body, motor vehicle and non-motor vehicle detection algorithms that can implement rich AI intelligent operations. And, this series of cameras are targeted and optimized for low-light capture, and the specific performance of the product is verified by a&s’ actual test.
Integrated bullet design
Looking from the exterior, the Zeno-Tech ZN-SNC-IB6200WS-I5ZPF (AO)-GLST network smart camera comes with an integrated bullet design, giving a simple and elegant look. The device measures 102 x 87 x 265 (mm) and weighs 800 grams. Although not too big, the camera’s smart applications are powerful. The device has two power supply options: DC12V (± 30 percent) and POE, as well as two ways of mounting: bracket and wall. As such, deployment is very simple, and installation is convenient. In addition, the local storage supports a Micro SD/SDHC/SDXC card (256G). The serial communication interface is configured with an RS485 interface. The reset port is equipped with a reset button. The network interface is equipped with an RJ45 10M/100M adaptive Ethernet port, and the interface protocol supports ONVIF/CGI/GB28181.
During the test, the performance of this device was quite stable and reliable. The working temperature range is -30°C to +70°C, the working humidity 10 to 90 percent. It is cold- and heat-resistant with an IP67 rating, and power consumption is small, only at 10W. Generally speaking, the hardware configuration of this camera is excellent, and its environment adaptability is strong. In addition, it supports Apple- and Android-based mobile phone monitoring, which is easy to use.
Highly accurate capture
The camera has built-in deep learning-based algorithms for faces, human bodies, motor vehicles and non-motor vehicles, which can realize all-target detection, capture and tracking. During the actual test, we first selected the “Human Face” function in the backend Snap Type setting to enable detection and capture of faces in motion. Results show high recognition sensitivity, allowing quick capture of optimal face image and relevant attribute information. Up to 30 faces were detected at the same time. Resolution-wise, the face capture capability is strong, with a minimum capture pixel of only 32 x 32. At the same time, tests show a face capture angle range of ±45°, and the accuracy of the face capture is more than 99 percent.
We then switched to the “FaceHumanNon-Motor” function on the backend to test detection of human bodies, motor vehicles and non-motor vehicles. The device can capture and extract attribute information with high accuracy and detect up to 70 humans/motor vehicles/non-motor vehicles at the same time. The minimum detection pixel of a motor vehicle is 40 x 40, and that of a human/non-motor vehicle is 32 x 64.
In terms of capture strategy, this device has the “best capturing” and “fastest capturing” modes to choose from, allowing the automatic upload of the original large image or the target cutout. It also supports pixel filtering, ROI area filtering, tracking ID filtering, quality score filtering and target de-duplication. Whether it is human face detection or human body-motor vehicle-non-motor vehicle full-target detection, the best image is output. In addition, this camera can single-handedly handle multi-dimensional perception and data extraction, which not only greatly reduces the complexity and cost of AI camera deployment, but also fully meets the full-target detection, capture and tracking needs at backstreets, alleys and community entrances.
Clear capture in low light
This camera by Zeno-Tech uses a 1/2.8" progressive scan CMOS sensor, with high-efficiency array infrared illumination – the IR distance is up to 50 meters. Smart IR is supported so that the brightness of infrared illumination is automatically adjusted to solve the near-target overexposure problem, ensuring brightness uniformity in images shot in IR. Tests show the minimum illumination of this device is 0.002Lux@F1.4 (AGC ON) and 0Lux@F1.4 (AGC ON) with IR, ensuring clear capture of faces and license plates even in ultra-low light environments. Meanwhile, image sharpness/brightness/saturation/contrast can be adjusted to increase image color originality.
In addition, the device supports day-and-night conversion, digital noise reduction, fog transmission function, wide dynamic range and other functions. Among them, the true wide dynamic range is 120dB. Combining it with the day-and-night conversion function allows the camera to shoot high-quality video even in environments with high contrast and unchanged lighting conditions. When testing the wide dynamic range and backlight compensation functions, the camera was pointed at a strong light source, and the color of the entire screen is relatively uniform. The background of the entire scene and the picture inside the illumination are displayed clearly, the image is clear, and faces and license plates are all clearly detailed.
Video smooth at 1080p
This network camera supports the industry's latest video coding standard, H.265, and is compatible with the existing mainstream H.264 standard. The maximum main stream is 1920 x 1080 @ 25/30fps, and the image is very clear. The device has a built-in 2.8-12mm electronically adjustable precision aperture lens, which can achieve the best monitoring effect according to the scene environment. During the test, the response from capturing to display is good, and the video playback is very smooth.
The camera has a motion detection function. At the backend, the user can set multiple detection areas and multi-level sensitivity and can select four areas for occlusion. During the test of the motion detection function, it was found the overall response was quite sensitive; if someone or a car passes by, it immediately triggers an alarm. There’s also a strong anti-interference function, which greatly reduces the false alarm rate, allowing the camera to flex its muscles in small scenes with a complex environment.
Features of product
1. Simple and fashionable integrated design, and excellent hardware configuration. Overall performance stable and reliable.
2. Employs advanced H.265/H.264 video coding algorithm with high compression ratio, realizing low bandwidth and high image quality.
3. Good low-light capture effect, ensuring capture of clear human faces and license plate information with high accuracy in a dark environment.
4. The combination of infrared illumination, day-and-night conversion and wide dynamic range allows clear color distinction and brightness uniformity in captured images.
5. Built-in multiple deep learning algorithms to achieve full-target detection and capture of the scene. The device has multiple uses, reducing the complexity and cost of camera deployment.
6. Working temperature range of -30°C to +70°C, working humidity of 10 to 90 percent, cold and heat resistance, high protection level, and strong environmental suitability.
Expert review
Zeno-Tech’s all-target detection network camera is part of the company’s AI IPC series. Building on the basic functions of traditional network cameras, the device includes built-in deep learning algorithms that can achieve highly accurate detection, tracking and capturing of human faces, human bodies, motor vehicles and non-motor vehicles. Whether it is the traditional video surveillance function or the AI function, the performance is very good. In backstreets, alleys of the community, internal roads, buildings, hotels, Internet cafes, restaurants, rural town roads, parks, bank offices and other smaller deployments where all-target detection and capturing are needed, the camera has obvious advantages.
More News