Integrating 24 Cameras into an Agricultural Machinery Retrofit

2026.10.27
Integrating 24 Cameras into an Agricultural Machinery Retrofit

A distributed camera acquisition and timing system integrated into an agricultural machinery retrofit, connecting 24 cameras to one NVIDIA Jetson AGX Orin controller.

*Based on a customer deployment. Customer details are omitted.

One AMR BOX and two Vision BOX units put 24 cameras on a single PTP time base, with two optical cables replacing 16 camera runs.

01 | The Challenge: 24 Cameras, Eight Direct Inputs

A customer retrofitting agricultural machinery for autonomous operation needed one controller to handle 24 cameras, alongside LiDAR, GNSS, an IMU and motor control. The AMR BOX controller, built around NVIDIA Jetson AGX Orin, provides eight direct GMSL camera inputs, leaving 16 additional cameras to be connected through another acquisition path.

Cameras needed24plus LiDAR, GNSS, IMU, motor control
Direct GMSL inputs8on the AMR BOX
Remaining16via another acquisition path

Connectivity was only half of the problem. The machine moves while it images the field, so frames captured at different instants show the scene from different positions. The cameras needed to capture in step with each other and share a time base with the rest of the machine's sensors.

02 | Distributing Camera Acquisition with Vision BOX

Albatron built the system from three units. The AMR BOX, an autonomous retrofit controller based on Jetson AGX Orin, connects eight GMSL cameras directly.

The other 16 cameras live in two Vision BOX units mounted at the front and rear of the machine. Each Vision BOX integrates eight MIPI cameras and an FPGA in a single enclosure, with the cameras wired to the FPGA over short internal cables. Each unit sends its camera data to the AMR BOX over its own 10G SFP+ fiber link, so two optical cables replace 16 individual camera runs across the machine.

Machine frame Vision BOX (front) FPGA + 8 MIPI cameras Short internal cables to the FPGA Vision BOX (rear) FPGA + 8 MIPI cameras Short internal cables to the FPGA 8 GMSL cameras Direct connection AMR BOX NVIDIA Jetson AGX Orin 2 × 10G SFP+ · built-in GNSS PTP grandmaster · CAN GMSL 10G SFP+ fiber 10G SFP+ fiber

03 | One Time Base, from GNSS to Every Camera

Timing starts at a u-blox F9P GNSS receiver built into the AMR BOX. The receiver supplies time-of-day information and a pulse-per-second (PPS) signal to the Jetson module, which serves as PTP/gPTP grandmaster (IEEE 1588 and 802.1AS) on the controller's Ethernet and fiber ports.

u-blox F9P GNSS receiver built into AMR BOX ToD + PPS Jetson module PTP/gPTP grandmaster IEEE 1588 / 802.1AS 8 GMSL cameras PPS to deserializers: frame sync, 30 Hz 16 MIPI cameras (2 Vision BOX) PTP to FPGA: triggers exposure start Network sensors, e.g. LiDAR PTP over 1G Ethernet Serial-time equipment NMEA GPRMC on M8 connector

From there, each group of devices is synchronized through the path that suits its interface:

Devices How they are synchronized
GMSL cameras PPS is routed to the deserializers, which generate frame sync for the eight directly connected cameras, at 30 Hz by default.
Vision BOX cameras Each Vision BOX synchronizes its FPGA clock with the AMR BOX over PTP. The FPGA uses this shared time base to trigger its eight MIPI cameras, aligning the start of exposure across the group.
Other devices PTP over the 1G Ethernet ports for network sensors such as LiDAR, plus a time-of-day output (NMEA GPRMC) on an M8 connector for equipment that accepts a serial time message.

Inside each Vision BOX, the FPGA handles both jobs: aggregating eight camera streams onto one optical link and timing their capture.

04 | What Albatron Delivered

Albatron designed and delivered the AMR BOX and both Vision BOX units as one system, including the timing chain that links them. The customer integrates the LiDAR and motor controllers into the machine through the AMR BOX's PTP-capable Ethernet and CAN interfaces.

AMR BOX
  • NVIDIA Jetson AGX Orin
  • 8 GMSL camera inputs
  • 2 × 10G SFP+
  • Built-in GNSS
  • PTP grandmaster
  • CAN
Vision BOX (×2)
  • 8 integrated MIPI cameras each
  • FPGA for camera aggregation and timing
  • 10G SFP+ uplink

05 | Applying the Architecture to Multi-Sensor Machinery

On a machine with cameras spread from front to rear, the camera interface decides where acquisition happens and how much cabling crosses the frame. Placing acquisition in units next to the cameras and linking them to the controller by fiber lets the camera count grow beyond the controller's direct inputs. Carrying PTP over the same fiber keeps every camera group on one time base.

Discuss your multi-sensor integration project

Albatron develops custom AI controllers and FPGA hardware for robotics, agricultural machinery and multi-sensor edge AI systems. Contact us about your camera connectivity, timing architecture and system integration requirements.

Contact us →