Artificial Intelligence Based Stereo Vision System

Name of demonstration

Main objective

Detect, recognize and classify randomly distributed objects that are overlapping each other in a pile and pick them up by a robotic arm.

Short description

A lot of industrial processes involve operation with a large number of different objects with an arbitrary location. It is hard to automate these kinds of processes because sometimes it is impossible to predetermine the positions for these objects. To overcome this issue, we integrate 3D and 2D computer vision solutions with AI and robotic systems for object detection, localization and classification.

Owner of the demonstrator

Institute of Electronics and Computer Science

Responsible person

Researcher Janis Arents janis.arents@edi.lv

NACE

C - Manufacturing

Keywords

Robotics, Manufacturing, Computer Vision, Artificial Intelligence, Bin-Picking.

Potential users

Tech integrators, SMEs

Benefits for the users

Implementing AI-based solutions can be effective in the manufacturing industry by improving process and product quality, reducing cycle time, decreasing costs and much more.

The use of AI-based stereo vision systems can reduce adjustment time, the complexity of the system and is more flexible to changes in manufacturing lines.

The AI-based stereo vision system of UC17 can be trained to detect and estimate the pose of different objects that are randomly distributed in a pile, therefore, enabling automation of industrial processes involving a different kind of objects with unpredictable positions.

Innovation

Traditionally working with randomly distributed objects requires human resources or dedicated sorting hardware that usually is spacious, expensive and hardly adjustable if product assortment changes. The problem becomes more complex if different kinds of objects are mixed in one pile and need to be sorted and structured into determined position and orientation. The proposed system deals with this kind of uncertainty of the environment by use of AI-based computer vision system that can detect and estimate the pose of such objects. In combination with an industrial robot detected objects can be picked up and placed in a determined position.

Risks and limitations

This use-case demonstration has been built upon the Robot Operating System (ROS). The compatibility with different hardware depends on the ROS driver availability of the intended hardware, such as Industrial Robots and 3D cameras. The system has been tested with ROS supported hardware: Universal Robots UR5 industrial robot and multiple RGBD cameras, such as Intel RealSense, Kinect and Zivid.