AI VisionCube S

The AI VisionCube S brings industrial-grade AI target recognition to drone payloads in a kit weighing under 200 grams. Built around a 1 TOPS image-processing board running a deeply-optimized YOLOv7 multi-target architecture, the AI VisionCube S identifies up to 50 dynamic targets per frame and locks onto human and vehicle subjects with sub-60ms analog video latency to the pilot’s goggles.

Pairing the AI processing board with a single short-focus visible-light camera (4mm focal length, 1/2.8″ CMOS, 69° horizontal field of view), the AI VisionCube S delivers 1920×1080 imagery at 30Hz with 7341 mV/lux·s sensitivity for low-light operations. Detection range covers vehicles out to 450 meters and human targets to 170 meters — sufficient for most short-to-mid-range tracking missions.

Onboard intelligent crosshair auto-lock pulls tracked subjects to the center of the frame in 0.5 seconds, while Kalman-based trajectory memory re-acquires briefly occluded targets within three seconds. The high-speed lock success rate reaches 92% in dynamic motion scenarios, dramatically reducing the pilot’s micro-adjustment workload during demanding flight operations.

Designed for plug-and-play integration, the AI VisionCube S accepts a 9–16V input, communicates via the CRSF protocol, supports BetaFlight and ArduPilot firmware, and integrates into 7-inch or larger industrial UAV airframes in under five minutes. Six remote-control channels manage target lock, snap-tracking, fine-adjustment, and flight-mode toggling. Custom-trained AI models can also be deployed for system integrators extending beyond the default human and vehicle target classes.