Image quality is crucial in video surveillance. Today, AI plays an increasingly important role in optimizing image quality. Camera's hardware components, such as systems-on-chip (SoC) and image signal processor (ISP), are also key in determining image quality.
Optimized by AI
Needless to say, AI is now used in cameras to process complex video data to achieve accurate object detection and classification. But more and more, AI is also tasked with image quality optimization, which is essential in video surveillance.
“AI algorithms help analyze scene content and adjust parameters like exposure, focus, and white balance in real time. With MOBOTIX ONE, our new AI-driven camera platform, these functions are becoming smarter and more adaptive. AI also enhances noise reduction and motion detection, enabling our cameras to dynamically respond to environmental changes and ensure consistent quality,” said Christian Cabirol, CTO of MOBOTIX.
Gerald BeomSeok Kim, Senior Product Manager at Hanwha Vision, uses as an example noise reduction, which becomes more effective with AI than with conventional methods.
“Conventional NR technology, which relies on mathematical filters, often blurs both noise and fine details. In contrast, AI-NR technology uses a deep learning model that accurately recognizes noise patterns. It selectively removes only the noise pixels while preserving fine details, resulting in a cleaner image and more reliable data for AI analytics. Clean, noise-free video data also contributes to enhanced bitrate efficiency,” he said.
Other hardware components
Meanwhile, hardware components such as the camera’s systems on chip and image signal processor also play a vital role in boosting image quality.
“The SoC and ISP are critical components in delivering high image quality. The ISP processes raw sensor data and applies enhancements like noise filtering, color correction, and dynamic range optimization,” Cabirol said.
“The SoC acts as the camera's brain, not just processing commands but also determining the overall efficiency and performance of the entire image processing pipeline,” Kim said. “The ISP is a core processor embedded within the SoC. It's responsible for converting the analog signals from the image sensor into a viewable digital image. The ISP's capabilities directly determine the visual quality of the video. Image quality features like NR, WDR, and distortion correction all depend on the ISP's processing power and algorithms for their performance.”
Advantages of in-house developed SoC
Camera vendors can acquire SoCs from third parties or make the chips on their own. An example of the latter is Hanwha Vision, which has rolled out their latest SoC, Wisenet 9.
According to Kim, the advantages of using an in-house developed SoC are particularly pronounced due to the following:
• Performance optimization: An in-house SoC is designed to be perfectly aligned with a company's cameras, software, and video processing algorithms. This allows for superior performance, lower power consumption, and the best possible image quality, which cannot be achieved with a generic, off-the-shelf chip.
• Technological flexibility and independence: Owning your own SoC frees you from reliance on external suppliers' product roadmaps. This allows a company to implement new technologies or enhance specific functions with greater flexibility. It also provides greater control over firmware and software, enabling the company to rapidly develop and deploy security patches and updates.
• Enhanced security: Security can be "baked in" at the hardware level from the SoC's design phase. This makes it possible to implement more robust security features like secure boot and hardware-based encryption.
Hanwha Vision leverages those advantages with Wisenet 9, which effectively improves image quality and AI analytics performance.
“The most distinctive feature of Wisenet 9 is its Dual NPU Architecture, which consists of two independent NPUs. One NPU is dedicated to AI-based image enhancement (including NR and WDR), while the other focuses exclusively on AI video analytics. This architecture ensures that complex image processing and analytics tasks can run at optimal efficiency simultaneously, without impacting each other's performance, delivering the best performance in real time,” Kim said.
Wisenet 9 also uses a hybrid architecture combining conventional NR with a more sophisticated AI-NR technology delivering vivid and sharp images without blurring or "hallucination" in low-light environments.
Kim adds that this exceptional noise reduction leads to dramatic efficiency gains. “A cleaner image requires less data to encode, allowing Wisenet 9 to effectively reduce bitrate,” he said.