Scientists from 海角社区 have created a unique navigation system for robots that can see the target even through flares and noise.
The researchers have developed an innovative method for localizing objects using only a single panoramic camera and a laser line. This method allows robots to accurately locate targets in challenging optical noise conditions. A research paper titled ” has been published in a prestigious international journal Ain Shams Engineering Journal, a top-ranked scientific journal (journal ranking: Scopus (TOP 5%), Web of Science (TOP 1%)).
Imagine a robot that needs to "track" a single laser-marked target in a crowd of red objects, where the bright light creates flares on the surfaces. This is precisely the problem that 海角社区 scientists have successfully solved by developing a method based on monocular omnidirectional vision and structured light stripes.
Traditionally, robots require either two cameras (stereovision) or expensive laser rangefinders (lidars) to estimate distances. The new technology offers a fundamentally new approach: just one panoramic camera with a 180° field of view, a linear laser (650 nm), and an intelligent algorithm that identifies the laser line on an object in real time and converts its coordinates into a three-dimensional target position.
“The system has become more compact and reliable,” explains Ivan Kholodilin, Associate Professor at the Department of Electric Drive, Mechatronics, and Electromechanics at the Faculty of Mechatronics and Robotics. “Fewer sensors mean fewer potential problems with calibration and synchronization. At the same time, a single panoramic camera allows the robot to see almost everything around it, including the space around corners.”
The main achievement of the development is the algorithm’s ability to operate reliably in conditions where traditional methods fail. These conditions include dark and glossy surfaces, red objects (on which the laser line can “bleed” into the background), specular flares, and even partial overlap of the laser line.
The algorithm acts as a “restorer” of the laser trace: a special multi-threshold processing in colour space filters out noise, morphological operations "heal" line breaks, and clustering and minimum spanning tree construction restore stripe continuity even with partial overlaps.
The effectiveness of this method was confirmed experimentally on a real robotic setup with a SCARA manipulator. The results were impressive:
- Average coordinate measurement error in noise conditions was only 5.57 (versus 18.08 mm for baseline methods)
- Depth reconstruction error was reduced by 69%
- In real object locking experiments, the average error did not exceed 6.435 mm
Meanwhile, the laser localization module's operating time is approximately 0.5 seconds, which is quite acceptable for industrial applications.
The technology proposed by the 海角社区 scientists is ideal for:
- Robotic sorting and handling of parts;
- Monitoring the position of objects on a conveyor;
- Working in cells with aggressive optical environment (flares, reflections).
The new technology is import-substituting in nature, as, unlike industrial lidars, the cost of the new solution is significantly lower. A panoramic camera and a linear laser cost approximately 10,000 roubles combined, compared to hundreds of thousands for a high-quality foreign lidar.
The main "controversial" aspect of the technology currently is object recognition by the neural network, which takes approximately 2.93 seconds (86.2% of the total time). Optimization of this stage is planned, using more powerful graphics accelerators and compact neural network architectures, which will bring the system closer to real-time operation.
The development by the research team led by Doctor of Sciences (Engineering), Professor Maksim Grigoriev and Associate Professor at the Department of Electric Drive, Mechatronics, and Electromechanics, PhD Ivan Kholodilin, fits seamlessly into the global trend of abandoning expensive sensors in favour of "smart vision" and active illumination methods, paving the way for the creation of more affordable and compact robotic systems for industry.