Engineered to maximize tracking area and deliver unmatched real-time large-area analytics, the PF-Series is purpose-built for modern convolutional neural networks (CNNs). It delivers breakthrough performance for CNN-based 3D vision workloads while providing the scalability needed for the next generation of AI models.
“Today marks a new era in retail analytics,” said Christian Studer, Xovis CEO and Co-Founder, speaking at NRF 2026 in Manhattan. “Xovis has put years of research and engineering into developing new sensor hardware with exceptional coverage and a powerful architecture. The research again proved that 3D stereovision outperforms other technology in responsibly measuring people flow in large environments.”
It delivers:
Up to 7.9 TOPS of CNN performance on an AI-native NPU: Accelerates the convolutional models that drive Xovis’ 3D tracking precision and optical pipeline.
NPU speeds up to 3× faster than the widely used NXP i.MX8 and up to 22.5× faster than PC2SE-Series: Establishes a strong foundation for more demanding CNN workloads.
Quad-core CPU (4×2.1 GHz) with 44.8 GFLOPS GPU: Delivers exceptional compute density for real-time processing and next-generation challenges.
AI-native architecture optimized for modern CNNs: Exceptional performance for CNN-based 3D vision processing while enabling future AI model innovations
“We’ve witnessed tremendous advances in compute capabilities in recent years. Xovis has transformed those into a market-ready sensor that supports retailers’ transition to smarter, more automated processes,” said Florian Eggenschwiler, Xovis CPO.
Designed as a balanced yet highly capable AI platform, the new SoC provides industry-leading CNN acceleration at a cost point typically associated with far less capable devices. By combining a maximized 3D optical system with a CNN-ready compute architecture, Xovis delivers a future-proof foundation for scalable, high-accuracy spatial analytics across retail, mobility, smart buildings, and broader smart-infrastructure applications.
