Ultralow-Power Edge AI Processor for Automotive Always-On Surveillance Systems

OMNIVISION is releasing the OAX7700, an ultralow-power application-specific integrated circuit (ASIC) engineered to function as an always-on vision companion processor. This hardware upgrades existing automotive exterior camera networks, such as surround-view systems, into AI-driven parking surveillance platforms capable of operating within the broader automotive data ecosystem.

Edge Computing for Parking Surveillance Systems
The OAX7700 integrates a neural processing unit (NPU) to shift computational burdens from central vehicle architectures directly to the edge. During always-on operation, the camera continuously monitors the immediate environment while maintaining a minimal power state. When a physical or visual event triggers the system, it instantly switches to standard operation, recording five to 10 seconds of pre-roll video. The vehicle's primary electronic control unit utilizes this footage to validate triggers and activate alarms, effectively eliminating false alerts caused by non-threat motions such as wind, debris, or foliage, while preserving continuous forensic evidence.

Power Efficiency and Algorithmic Customization
By processing human presence detection, object classification, and distance measurement on the edge, the ASIC reduces power consumption by 97% compared to existing surveillance architectures on the market. The built-in NPU permits original equipment manufacturers to develop and deploy proprietary machine learning algorithms directly onto the chipset. To support the transition toward software-defined vehicles, the processor allows for over-the-air functional updates, preserving the ASIC's low power consumption while expanding operational capabilities.

Hardware Footprint and Production Timeline
To achieve a compact physical footprint suitable for exterior cameras, the processor utilizes stacked PSRAM and flash memory enclosed in a 60-pin BGA package measuring 5.5 mm by 3.5 mm. David Ho, senior marketing manager at OMNIVISION, noted that the component allows manufacturers to deploy parking surveillance functionality that can be updated throughout the vehicle's lifetime, utilizing established architectural principles from the security market segment.

The hardware was showcased at AutoSens Europe 2026 in Booth 100 (September 22-24) in Barcelona, Spain.

Additional Context
This section details technical specifications and competitive benchmarking not included in the original news release.

In the automotive edge AI market, companion processors like the OAX7700 compete with low-power vision processing units such as the Ambarella CVflow series and NXP Semiconductors' S32V vision processors. The standard benchmarking metric for this class of hardware is the performance-per-watt ratio (TOPS/W) and standby power consumption (typically measured in milliwatts during sleep or monitoring modes). Devices in this category rely on dedicated hardware accelerators to process localized convolutional neural networks without waking the primary central vehicle compute platforms, such as Qualcomm Snapdragon or NVIDIA DRIVE systems. Furthermore, integrating stacked PSRAM directly into a BGA package is an established hardware mechanism designed to reduce memory bandwidth bottlenecks while strictly limiting the physical space required inside highly constrained exterior camera modules.

Edited by Aishwarya Mambet, Induportals Editor, with AI assistance.

www.ovt.com