www.ptreview.co.uk
01
'26
Written on Modified on
Edge AI Vision System Architecture and MIPI Camera Integration
Vision Components introduces an intelligent board-level camera platform utilizing onboard processing and high-resolution image sensors for industrial machine vision applications.
www.vision-components.com
Vision Components presents the VC EvoCam, an all-in-one intelligent board-level camera with MediaTek processor for even faster and easier embedded vision integration, at CVPR 2026.
Vision Components has developed a highly integrated board-level camera system engineered for embedded vision and edge AI workloads. Measuring 65 by 40 millimeters, the hardware consolidates image acquisition and processing into a single module, eliminating the need for external computing units. The platform targets computer vision tasks across industrial automation, robotics, and Internet of Things architectures. Vision Components exhibits the hardware at the Conference on Computer Vision and Pattern Recognition (CVPR) from June 3 to 7 in Denver, Colorado.
Processing Capabilities and Hardware Interface
The core of the system utilizes a MediaTek Genio 510 Edge AI processor. This compute unit features a heterogeneous architecture comprising two ARM Cortex-A78 cores and four ARM Cortex-A55 cores, alongside an ARM Mali GPU. Neural network workloads are handled by an integrated NPU delivering a performance of 3.2 TOPS. The module is equipped with up to 2 GB of RAM and 16 GB of flash memory, with expandability options via an SD 3.0 interface to manage extensive image data processing and storage. Operating on a customized Debian Linux distribution, the system supports standard image processing functions directly.
Hardware integration relies on a 100-pin board-to-board connector that exposes critical processor interfaces, including I/O, I²C, USB, Ethernet, Video DSI, and PCIe. The system architecture supports both onboard image sensors and cable-connected remote-head configurations. Initial iterations feature the Sony IMX900 image sensor, providing a 3.2-megapixel resolution and global shutter capability. To facilitate device integration, a baseline interface board providing power, trigger I/O, USB, and RJ45 LAN will be available for volume production in the third quarter of 2026, followed by a comprehensive development kit routing all connector signals to physical interfaces.
High-Resolution Global and Rolling Shutter Integration
Expanding the MIPI Camera portfolio, the new MIPI IMX540 module implements a global shutter mechanism using a Sony Pregius S series sensor. Operating in a 1.2-inch format with a 2.74-micrometer pixel size, the sensor achieves a 24.5-megapixel resolution (5,328 by 4,608 pixels) at 22 frames per second in 8-bit mode. The sensitivity of this architecture supports computer vision applications requiring precise object detection, such as humanoid robotics and complex machine vision inspection.
For applications necessitating sensitivity extending into the near-infrared range, the MIPI AR2020 module utilizes an Onsemi Hyperlux LP rolling shutter sensor. This 20-megapixel module captures images at a resolution of 5,120 by 3,840 pixels, operating at up to 24 frames per second in 10-bit capture mode. The sensor's low power consumption characteristics make it suitable for distributed edge AI and industrial IoT network deployments.
The new VC MIPI IMX540 global shutter camera module offers a resolution of 24.5 megapixels and is suitable for a wide range of applications in AI and machine vision.
Signal Routing and Protocol Implementation
To ensure broad compatibility, Vision Components provides open-source drivers for its portfolio of over 50 image sensors. The underlying hardware ecosystem includes components for prototyping and series integration, such as micro-coaxial and GMSL2 cabling capable of routing signals up to 10 meters. Furthermore, FPGA accelerators allow for hardware-level data processing directly within the MIPI data stream before the signal reaches the primary host processor. The integration of high-compute processors and open Linux operating systems represents a continuation of the industrial smart camera architecture initially developed by Vision Components founder Michael Engel in 1996.
Additional Context: This section details technical specifications and competitive benchmarking not included in the original product announcement.
The integration of a 3.2 TOPS NPU in the MediaTek Genio 510 positions the system competitively within the mid-range edge AI computing segment. Comparable embedded vision processors, such as the NXP i.MX8M Plus, typically offer around 2.3 TOPS, while higher-tier modules like the Rockchip RK3588 provide up to 6 TOPS. Operating a 24.5-megapixel sensor at 22 frames per second requires substantial data throughput, approaching 538 megapixels per second. This necessitates highly optimized utilization of the MIPI CSI-2 4-lane interface, which possesses a theoretical bandwidth limit of approximately 10 gigabits per second. Effectively managing this data volume at the edge relies on the integrated processing pipeline and NPU to execute localized inference tasks, thereby preventing transmission bottlenecks in industrial vision networks.
Edited by an industrial journalist, Lekshman Ramdas, with AI assistance.
Vision Components has developed a highly integrated board-level camera system engineered for embedded vision and edge AI workloads. Measuring 65 by 40 millimeters, the hardware consolidates image acquisition and processing into a single module, eliminating the need for external computing units. The platform targets computer vision tasks across industrial automation, robotics, and Internet of Things architectures. Vision Components exhibits the hardware at the Conference on Computer Vision and Pattern Recognition (CVPR) from June 3 to 7 in Denver, Colorado.
Processing Capabilities and Hardware Interface
The core of the system utilizes a MediaTek Genio 510 Edge AI processor. This compute unit features a heterogeneous architecture comprising two ARM Cortex-A78 cores and four ARM Cortex-A55 cores, alongside an ARM Mali GPU. Neural network workloads are handled by an integrated NPU delivering a performance of 3.2 TOPS. The module is equipped with up to 2 GB of RAM and 16 GB of flash memory, with expandability options via an SD 3.0 interface to manage extensive image data processing and storage. Operating on a customized Debian Linux distribution, the system supports standard image processing functions directly.
Hardware integration relies on a 100-pin board-to-board connector that exposes critical processor interfaces, including I/O, I²C, USB, Ethernet, Video DSI, and PCIe. The system architecture supports both onboard image sensors and cable-connected remote-head configurations. Initial iterations feature the Sony IMX900 image sensor, providing a 3.2-megapixel resolution and global shutter capability. To facilitate device integration, a baseline interface board providing power, trigger I/O, USB, and RJ45 LAN will be available for volume production in the third quarter of 2026, followed by a comprehensive development kit routing all connector signals to physical interfaces.
High-Resolution Global and Rolling Shutter Integration
Expanding the MIPI Camera portfolio, the new MIPI IMX540 module implements a global shutter mechanism using a Sony Pregius S series sensor. Operating in a 1.2-inch format with a 2.74-micrometer pixel size, the sensor achieves a 24.5-megapixel resolution (5,328 by 4,608 pixels) at 22 frames per second in 8-bit mode. The sensitivity of this architecture supports computer vision applications requiring precise object detection, such as humanoid robotics and complex machine vision inspection.
For applications necessitating sensitivity extending into the near-infrared range, the MIPI AR2020 module utilizes an Onsemi Hyperlux LP rolling shutter sensor. This 20-megapixel module captures images at a resolution of 5,120 by 3,840 pixels, operating at up to 24 frames per second in 10-bit capture mode. The sensor's low power consumption characteristics make it suitable for distributed edge AI and industrial IoT network deployments.
The new VC MIPI IMX540 global shutter camera module offers a resolution of 24.5 megapixels and is suitable for a wide range of applications in AI and machine vision.
Signal Routing and Protocol Implementation
To ensure broad compatibility, Vision Components provides open-source drivers for its portfolio of over 50 image sensors. The underlying hardware ecosystem includes components for prototyping and series integration, such as micro-coaxial and GMSL2 cabling capable of routing signals up to 10 meters. Furthermore, FPGA accelerators allow for hardware-level data processing directly within the MIPI data stream before the signal reaches the primary host processor. The integration of high-compute processors and open Linux operating systems represents a continuation of the industrial smart camera architecture initially developed by Vision Components founder Michael Engel in 1996.
Additional Context: This section details technical specifications and competitive benchmarking not included in the original product announcement.
The integration of a 3.2 TOPS NPU in the MediaTek Genio 510 positions the system competitively within the mid-range edge AI computing segment. Comparable embedded vision processors, such as the NXP i.MX8M Plus, typically offer around 2.3 TOPS, while higher-tier modules like the Rockchip RK3588 provide up to 6 TOPS. Operating a 24.5-megapixel sensor at 22 frames per second requires substantial data throughput, approaching 538 megapixels per second. This necessitates highly optimized utilization of the MIPI CSI-2 4-lane interface, which possesses a theoretical bandwidth limit of approximately 10 gigabits per second. Effectively managing this data volume at the edge relies on the integrated processing pipeline and NPU to execute localized inference tasks, thereby preventing transmission bottlenecks in industrial vision networks.
Edited by an industrial journalist, Lekshman Ramdas, with AI assistance.
