“In UNIKIE we developed multi-LiDAR fusion architecture, aimed at real-time scanning of large indoor or outdoor volumes, allowing detection, tracking and analysis of the dynamically-moving objects with the required level of accuracy and robustness.”
In many environmental scanning applications, the single LiDAR sensor is usually sufficient for achieving high-quality results. In these scenarios, long continuous recording is performed, in order to obtain large number of scans. These scans can later be aligned and merged together to obtain reach and highly detailed 3D model of the scene.
In the real-time applications, however, where fast environmental changes should be detected and analyzed, single LiDAR sensor may not be sufficient. Thus, the quality of the obtained scans can only be only improved by utilizing several synchronous sensors, accompanied by real-time merging technique.
Following the same reasoning, majority of autonomous car manufacturers equip their self-driving prototypes with multiple LiDARs and other sensors. In those automotive cases, however, all the sensors are rigidly mounted and restricted to be relatively close to each other, which leads to fast decay of the scanning density with the distance.
In contrast, for a number of applications scanning density should be kept constant over some desired “scanning volume”. These applications for instance include indoors and outdoors surveillance, motor-traffic tracking and counting, analysis of human behavior (e.g. pedestrians/visitors/patients), among other things.
In Unikie we developed multi-LiDAR fusion architecture, aimed at real-time scanning of large indoor or outdoor volumes, allowing detection, tracking and analysis of the dynamically-moving objects with the required level of accuracy and robustness. LiDAR sensors can be arranged in a way to guarantee efficient detection even in case of highly-occluded or over-populated environment.
Figure 1: Difference illustrated between a) single-LiDAR scanning and b) using dual LiDARs.
On the figure 1 difference between single-LiDAR scanning (left) and scanning with two LiDARs (right) is illustrated. Sensor positions and orientations are shown with Red-Green-Blue coordinate systems. Here, the two-LiDAR approach managed to capture and detect an object looking as a camera-tripod, which was occluded otherwise. Additionally, all the objects (and subjects) are captured with much higher density, which significantly improves temporal tracking and consequently, the overall system performance.
In the following example (see figure 2), two persons, standing close to each other were detected as a single cluster in the single-LiDAR system. Nevertheless, they were perfectly distinguished with the two LiDARs. Here again, increased density with two LiDARs provides much more data for accurate and robust detection and analysis.
Figure 2: Detecting close but separate objects using two LiDARs providing more accurate data.
– I strongly believe that developed technology for real-time multi-LiDAR fusion can find applications and prove its usefulness in many other areas and problems, says Sergey Smirnov, LiDAR expert in Unikie R&D team.