Depth sensors

Depth cameras: what are they used for?

Depth cameras are used in many mobile robots (and autonomous cars) to assess the robot's surroundings with a great accuracy. They can also be used for precision gripping tasks (medical, high-tech industry).

Depth cameras (or 3D cameras) are able to work inside and outside, in a wide range of lighting conditions.

Constraints of mobile robotics

Autonomous mobile robots are often required to move in dynamic environments (stores, hospitals, airports, etc.). The robot must be able to slow down depending on the density of the crowd, stop at the last moment if facing an obstacle, and find the most appropriate path on its own.

A dynamic environment also implies that items can be often moved around (construction sites, industrial sites, etc.). Hence, mobile robots must be highly responsive (in real time) and extremely accurate.

Taking the time to select the sensors that will be fitted on your AMR is therefore a key factor in any robotics project.

The 3 most common sensors used in robotic vision

For vision, the sensors chosen are often :

• Depth cameras
• LiDAR
• Ultrasonic proximity sensors

These sensors are complementary and offer the robot a highly accurate sense of its environment.

Depth cameras

They render what human eyes see in the most reliable way possible: they distinguish shapes, colors and movements. They have a wide angle view. The stereo image detection technology uses two cameras to calculate the depth of the image (thanks to the merging of the two images received separately).

These cameras tend to have false positives or "ghost". Other sensors, such as LiDAR or ultrasonic sensors, are often use to enhance or correct the camera's data collection.

LiDAR

LiDAR measures the distance between itself and a target. It will be able to recreate in real time a 3D map of its surroundings (point cloud). It is a fast and accurate technology, however, LiDARs cannot detect any glass surfaces or mirrors.

Ultrasonic proximity sensors

Ultrasonic proximity sensors can detect glass and mirror.

Generation Robots, distributor of Intel RealSense and Stereolabs cameras

Stereolabs cameras: high performance sensors

ZED cameras are passive depth sensors. This means that they do not emit laser or IR light like active sensors. Several passive sensors can be used at the same time without any kind of interference. They are also not affected by sunlight.

ZED stereo cameras have a high resolution and frame rate compared to active sensors. They have a longer depth range than active sensors.

Intel Realsense cameras: good performance + GPU

The big asset of Intel Realsense cameras is that they are equipped with a graphics card. No need to use a computer that has a GPU to make it work.

A NUC or even a Raspberry Pi can do the job, which can help lower the hardware costs for your project.

Luxonis cameras: robotics accessible to all

Luxonis cameras are turnkey products with the ability to run a first script in less than 30 seconds. They integrate AI, computer vision and image processing directly on the device. All cameras are based on Depth AI, an open-source software suite.

The brand markets a number of models with the option of choosing between the basic version or the PoE version. The PoE option is a more robust version with a waterproof housing that allows it to be used outdoors or in more hostile environments.