At the current stage, the resolution of intelligent driving cameras is steadily increasing.
A single camera moving from several million pixels to over ten million pixels will certainly produce finer images. However, for intelligent driving, a clearer image does not necessarily mean more useful information has been obtained.
In 2023, Sony Semiconductor Solutions released the automotive image sensor IMX735, featuring approximately 17.42 million effective pixels and emphasizing the role of high resolution in capturing distant targets.
In 2025, Sony Semiconductor Solutions released another automotive image sensor, the IMX828, with approximately 8.34 million effective pixels. This product shifted its design focus from extreme resolution to comprehensive perception performance, simultaneously improving HDR (High Dynamic Range), low-light performance, and motion blur suppression capabilities, achieving a dynamic range of up to 150 dB.
Why did the pixel count decrease in 2025? What are intelligent driving cameras truly pursuing?
01. What Does Increasing Pixels Actually Solve?
The most significant change brought by higher camera resolution is the ability to capture more pixels within a unit field of view.
This is crucial for distant targets.
Assuming the same camera field of view and lens remain unchanged, a distant vehicle may occupy only a few pixels in a low-resolution image. A high-resolution sensor allows this target to occupy more pixels in the image, thereby preserving details such as outlines, headlights, and road signs. However, there is a prerequisite: these additional pixels must contain valid information.
If the lens, sensor sensitivity, and imaging environment do not improve synchronously, merely increasing the spatial sampling density of the image will not result in a proportional increase in valid information for the intelligent driving model.
Especially for distant targets, whether a target can be clearly seen depends not only on the number of pixels but also on the lens field of view angle, focal length, sensor size, pixel size, exposure conditions, and signal-to-noise ratio.
In other words, resolution determines how finely space can be sampled, but it cannot independently ensure that these samples are clean and stable enough.
However, blindly pursuing higher resolution is also problematic. More pixels lead to greater data throughput and heavier computational burdens on the backend; in systems with limited computing power, this may cause processing delays due to inability to keep up, thereby affecting the real-time performance and stability of perception.
That is to say, while low resolution makes it difficult to see distant targets clearly, stacking too high resolution brings issues such as reduced light sensitivity per pixel and increased computational load.
This is why current automotive image sensors do not simply pursue pixel count but instead enhance dynamic range, low-light capability, and high-speed imaging capability alongside resolution.
02. Why Might HDR and Low-Light Performance Be More Important Than a Few Million Extra Pixels?
Lighting conditions in real-world road environments are highly complex.
Vehicles may just have exited an underground garage, driven into strong backlight at noon, or entered a tunnel facing scenes with drastic changes in brightness and darkness.
In such cases, even if the camera has very high resolution, if bright areas are already overexposed and dark areas have lost detail, adding more pixels cannot retrieve the already lost information.
This is the problem that High Dynamic Range aims to solve: enabling the camera to retain information from both bright and dark areas within the same scene.
For example, at a tunnel exit, the sunlight area is very bright, while the interior of the tunnel remains relatively dark. If exposure prioritizes the bright areas, the dark areas easily turn completely black; if it prioritizes the dark areas, severe overexposure may occur near the exit.
Automotive HDR works by expanding the range of brightness levels the sensor can record simultaneously.
Sony's publicly disclosed automotive HDR technology uses pixels with different characteristics or different exposure methods to obtain a wider dynamic range, reducing overexposure and underexposure in scenarios such as backlight and tunnel entrances/exits. Low-light performance faces similar challenges.
On nighttime roads, a pedestrian may reflect very little light.
If the sensor's inherent light sensitivity is insufficient, even increasing the output resolution from 4 million to 8 million pixels may still result in dark areas filled with noise.
For intelligent driving, such noise can obscure target edges, making it difficult for models to determine what the target actually is.
Therefore, high resolution increases spatial information, while HDR and low-light capabilities improve the validity of that information.
The two address different problems.
03. Why Is Seeing Clearly While Moving Another Matter?
Intelligent driving cameras typically operate while the vehicle is in motion.
When a car travels at high speed, the camera needs to continuously acquire road images. If the exposure time is long, displacement of the vehicle and targets during exposure may cause motion blur.
At this point, pedestrians, vehicle edges, or road signs that should originally be clear become blurred in the image.
This means that what intelligent driving requires is not a static high-resolution photo, but a continuous, stable image sequence suitable for computer vision recognition.
This is also why automotive image sensors have begun to pay attention to high-speed readout, HDR, and motion artifacts simultaneously.
There is another easily overlooked issue: higher HDR is not always better.
When synthesizing information from different exposures, if objects in the frame are moving rapidly, the time difference between exposures may produce motion artifacts. Therefore, automotive HDR requires balancing dynamic range, low-light performance, motion handling, and readout speed.
This is why discussing an intelligent driving camera based solely on megapixel count is no longer sufficient.
What truly determines how much valid information it provides to the intelligent driving model is the imaging quality formed jointly by resolution, optical system, light sensitivity, dynamic range, exposure, and readout speed.
From this perspective, the development of cameras from high pixel counts to even higher pixel counts is not wrong. Distant targets, road signs, and small objects indeed require sufficient spatial resolution.
However, if the new pixels do not simultaneously receive sufficient lighting information, and if backlight, night noise, and motion blur are not resolved, then what increases is primarily the data volume, not necessarily the visual information that intelligent driving truly needs.
Thus, what intelligent driving cameras truly pursue is not a clearer photo after zooming in, but consistently obtaining sufficiently stable, complete, and effectively usable visual information amidst high-speed vehicle movement, day-night variations, backlight, and complex road environments.
This is also why today's competition in automotive cameras is no longer just about pixel count.