Sony Patent Uses Camera Images to Verify and Boost Depth Sensor Reliability
Depth sensors can lie, and Sony wants a way to catch them in the act. This patent describes a system that measures the same distance two different ways, then scores which reading to trust.
What Sony's depth-sensor confidence check actually does
Imagine a robot or a self-driving camera trying to judge how far away a wall is. It has two tools: a depth sensor (like the kind that shoots invisible laser pulses) and a regular camera. Sometimes the depth sensor gives a bad reading, maybe because of bright sunlight, a reflective surface, or just sensor noise. The problem is the device has no easy way to know when that's happening.
Sony's patent describes a solution: calculate the distance to that same surface independently using just the camera image, then compare the two numbers. If they match closely, the depth reading is probably good. If they're far apart, something is off and the system should trust that depth data less.
The result is a confidence score attached to each depth reading, so whatever software is using that data, whether it's a drone, a camera, or an AR headset, knows how much weight to put on it.
How the two distance calculations get compared
The system has four main components working together:
- Acquisition section: pulls in both a camera image and depth data from the same sensor area at the same time.
- First calculation section: uses the camera image alone to estimate where the sensor is positioned in space (a technique called visual odometry or camera-based pose estimation, which figures out location from how objects appear in a photo), then calculates a distance from that estimated position to the scene.
- Second calculation section: calculates the distance using the raw depth data directly, the kind a LiDAR or time-of-flight sensor produces.
- Confidence calculation section: compares the two distances and produces a confidence score for the depth data.
The core idea is that a camera image and a depth sensor are independent sources of truth about the same physical space. When they agree, you can trust the depth reading. When they disagree significantly, something has degraded the depth data and downstream systems should treat it with skepticism.
The patent doesn't specify a fixed formula for scoring confidence, leaving room for the gap between the two distances to be weighted in different ways depending on the application.
What this means for robots, AR, and self-driving cameras
Depth sensors are central to a growing list of devices: AR headsets, autonomous vehicles, delivery drones, and 3D cameras. All of them suffer from the same blind spot, they can't easily tell when their own depth readings are wrong. A bad depth measurement fed into navigation software can cause a robot to misjudge an obstacle, or an AR headset to place a virtual object in the wrong spot.
By attaching a reliability score to each depth reading in real time, this approach lets devices automatically down-weight bad data rather than act on it blindly. For you as an end user, that could mean more stable AR overlays, fewer navigation errors, or a camera that handles tricky lighting conditions more gracefully.
This is a solid, practical patent rather than a flashy one. Cross-validating sensors against each other is a well-understood engineering principle in robotics and autonomous systems, but packaging it as a dedicated confidence-scoring layer for depth data is a useful building block. Sony's camera and sensor hardware business gives this real commercial relevance, particularly as depth sensors show up in more consumer devices.
Get one Big Tech patent every Sunday
Plain English, intelligent commentary, no hype. Free.
Editorial commentary on a publicly published patent application. Not legal advice.