Newshot Shanghai HiSilicon launches CS6286 image sensor, extending dashcam sentry system battery life to over 7 days
IT之家 November 13 news: On November 12, Shanghai HiSilicon launched the all-new 8-megapixel image sensor CS6286, which has now entered mass production.
According to the official introduction, it innovatively adopts a "sensing-computing-storage integrated" architecture, featuring core technologies such as high-precision intelligent analysis on the sensing side (CIS chip), low-power pre-recording, and all-scenario high dynamic range imaging, providing dashcams with a brand-new "all-weather intelligent sentry" solution that meets the stringent demands of smart mobility, in-vehicle safety, and other fields for long battery life, high image quality, and high reliability.
The CS6286 adopts a 1/1.8" optical format and a 2.0μm pixel size design, provides 8-megapixel (3840×2160) physical resolution, and supports native 4K@30fps HDR video recording.
Under traditional aftermarket dashcam solutions, sentry systems typically support only about 24–48 hours of recording. The CS6286 is equipped with innovative sensing-side AI technology that, through two major mechanisms—"AI precise wake-up + pre-recording power saving"—extends the sentry system's battery life to more than 7 days. AI wake-up is responsible for "working only when there is a real threat," while pre-recording power saving is responsible for "consuming as little power as possible during normal times"; together they achieve a revolutionary change in energy consumption patterns.
The CS6286 integrates a lightweight AI engine with equivalent 50GOPS@int8 of computing power, capable of rapidly processing image data on the sensing side and completing intelligent analysis of specific tasks: in common scenarios such as parking lots, roadsides, and courtyards, the detection accuracy for humanoid targets reaches over 95%.
In addition, the CS6286 is delivered with a complete AI training suite for degradation, retraining, and compilation, supporting custom categories with up to 3 target classifications, meeting customers' needs for differentiated scenario competitiveness. Users input custom datasets and corresponding parameters, and the tool outputs a usable model parameter configuration file.
After the vehicle is turned off, the main control SoC is powered down, and the system runs solely on the CS6286 single chip at extremely low power consumption, completing pre-recording, storage, and intelligent recognition.
It uses built-in SRAM to cyclically store the most recent 10 frames of VGA resolution (640×432) images, while keeping the sensing-side AI engine continuously on low-power watch to check for environmental risks. Once the AI engine confirms a valid target, the sensor can wake up the main control SoC through a dedicated wake-up channel within 50 milliseconds, while simultaneously saving the buffered frames from the 10 frames before the trigger.
