Rockchip RV1106 및 6축 자이로 EIS 기반 반려동물 1인칭 POV 멀티모달 AI 비전 목걸이

Client

A global pet social media agency and lifestyle smart hardware brand focused on unlocking companion animals' perspectives through AI wearables and creator content ecosystems.

Challenge

Pets run, sprint, and shake with extreme angular velocities, turning unassisted action camera footage into an unwatchable, nauseating blur. Sifting through 8+ hours of daily continuous footage imposes a severe manual editing bottleneck. Furthermore, intense video encode thermal dissipation inside a compact 45g collar poses a safety risk to animal skin.

Product Concept & Architecture

The system utilizes the Rockchip RV1106G2 edge vision processor, integrating an efficient Cortex-A7 CPU, 0.5 TOPS NPU, and dedicated ISP 3.0. Paired with a Sony IMX335 5MP STARVIS low-light image sensor and a custom 140° ultra-wide low-distortion lens assembly. Stabilization is driven by a Bosch BMI270 6-axis IMU tightly synced with the ISP pipeline, supported by 32GB eMMC storage and dual-band Wi-Fi 6. Thermal dissipation employs aerospace-grade phase-change graphite layers spreading heat evenly across the housing.

Engineering Solution

1. 6-Axis IMU-ISP Hardware EIS Stabilization: The BMI270 samples motion at 250Hz. A hardware quaternion trajectory filter syncs frame-by-frame with ISP roll/pitch vectors, completely eradicating dizzying pet gait wobble to output cinematic 2K 30fps HDR video.
2. Edge NPU Multi-Modal Highlight Triggering: A quantized INT8 convolutional vision-motion network runs continuously on the 0.5 TOPS NPU, identifying behavioral peaks: chase sprints, sniffing playmates, owner greetings, and feeding times.
3. Automated AI Video Diary Generation: When docked to its charging cradle, the collar offloads curated clips over Wi-Fi 6. A multimodal LLM automatically edits, captions, and scores a 60-second daily highlight vlog narrated from the pet's first-person voice.

Result

Video completion rates on social platforms surged by 340% compared to unstabilized action cams. Automated AI clipping achieves a 92% user retention rate, delivering 4.5 hours of continuous capture with surface chassis temperatures strictly below 38°C.