Intersection safety places demanding requirements on automotive perception systems. A vehicle turning across an intersection must detect pedestrians and cyclists within a wide horizontal area, while also maintaining the ability to identify vehicles farther ahead. Hitachi Automotive Systems announced a stereo camera AEB development intended to address both needs using camera-based sensing, without millimeter-wave radar.
The design is notable for its approach to the camera’s field of view and image processing. Rather than relying on one detection method across the entire scene, it uses different methods in the center and peripheral portions of the detection range. For engineers and system integrators, this illustrates how sensor architecture and processing strategy can work together to meet competing requirements: broad coverage near the vehicle and continued detection at distance.
Why intersection AEB needs a wider view
Automatic emergency braking (AEB) systems are intended to help prevent or mitigate collisions by responding to detected hazards. At intersections, the sensing challenge changes as a vehicle approaches or makes a turn. A system needs to observe a wide area to identify pedestrians and other vulnerable road users who may enter the vehicle’s path, while still monitoring vehicles ahead.
These requirements can create a design trade-off. Expanding horizontal coverage can help capture objects toward the sides of the vehicle, but the system must also preserve useful detection of distant objects. The announcement describes the camera as combining a wider-angle view with long-range detection to support intersection braking while retaining distant-object detection used for adaptive cruise control (ACC).
At the time of the announcement, the company also pointed to the growing emphasis on AEB in vehicle safety assessments, including its planned inclusion among NCAP assessment criteria from 2020. That context highlights why perception systems must account for more than straightforward forward driving: intersection scenarios require detection across a broader field.
How the stereo camera expands detection
The system widens the stereo camera’s horizontal detection range compared with conventional stereo cameras. It does this by shifting images detected by the left and right cameras. According to Hitachi Automotive Systems, this method provides an angle of view three times wider than that of conventional stereo cameras, without requiring a large increase in the number of sensor pixels.
Stereo cameras use images from two camera viewpoints to support depth perception. In this design, that binocular approach is used centrally, while image-processing methods are adapted for the periphery. The aim is to use the strengths of different techniques across the camera’s expanded detection area, rather than applying the same process uniformly.
Different processing for the center and periphery
The central region uses binocular detection, drawing on the two-camera view. In the peripheral region, the system uses monocular vision and time-series image processing to detect three-dimensional objects. The announcement describes these methods as being integrated with stored identification patterns developed using machine-learning methods.
This combination is intended to support accurate detection and range estimation of pedestrians and cyclists. The distinction between center and periphery is important: widening the view is not only a matter of capturing more of the scene. The system also needs to interpret objects across that scene and estimate their position and distance in a way that supports vehicle safety functions.
Hitachi Automotive Systems developed the technology for changing detection methods across the camera’s central and peripheral areas in collaboration with the Hitachi research and development group. The integrated approach aims to provide intersection detection using the stereo camera alone, reducing the need for additional sensing components such as millimeter-wave radar in this application.
Implications for perception-system design
For automotive engineers, the development underscores several practical considerations in camera-based perception:
- Coverage and range must be designed together. Intersection functions call for a wide horizontal field, while forward-driving functions may depend on detecting objects farther away.
- Processing can vary by region. Binocular detection in the center and monocular, time-series processing in the periphery allow the system to apply different techniques across its field of view.
- Sensor count is only one part of system design. The described approach seeks to broaden detection without a large increase in sensor pixels and without relying on millimeter-wave radar for the stated function.
- Detection must support useful estimates. Identifying a pedestrian or cyclist is not the only requirement; estimating the object’s range is also relevant to a safety response.
These points are relevant beyond a single camera implementation. In perception-system development, engineers must consider how sensor geometry, image processing, object identification, and range estimation combine to serve a particular driving scenario. The system described by Hitachi is an example of addressing that challenge through coordinated optical and processing choices.
Conclusion
Hitachi Automotive Systems’ stereo camera development targets a specific but demanding safety use case: automatic emergency braking while turning at intersections. Its wider horizontal view, region-specific detection methods, and continued distant-object detection are intended to support that function using a stereo camera without millimeter-wave radar. For technical decision-makers, the central takeaway is that broader coverage and reliable perception depend not just on the sensor’s view, but also on how image data is processed across that view.