DE202022104107U1 - Device for detecting objects - Google Patents

Abstract

Device (10, 90) for detecting at least one object (22, 24, 96) comprising a plurality of object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5) in a monitoring area (16, 100a, 100b) with
- at least one FMCW LIDAR sensor (12, 92a, 92b) for emitting transmitted light beams (14, ..., 14.n, 102.1, ..., 102.n) into the monitoring area (16, 100a, 100b) , for scanning a large number of measuring points (20.1, ..., 20n, 104.1, ..., 104.n) and for generating measurement data (18) from the measuring points (20.1, ..., 20n, 104.1, . .., 104.n) remitted or reflected transmitted light, the measurement data (18) comprising radial velocities (v ^r _n ) of the measuring points (20.1, ..., 20n, 104.1, ..., 104.n),
- a control and evaluation unit (32) for evaluating the measurement data (18), the control and evaluation unit (32) being designed to determine the measuring points (20.1, ..., 20n, 104.1, ..., 104.n ) to segment and at least partially assign them to the object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5), characterized in that the control and evaluation unit (32) is further designed to use the radial velocities (v ^r _n ) of the measuring points assigned to the object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5), a movement pattern of at least one of the object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5).

Description

The invention relates to a device for detecting objects according to the preamble of claim 1.

Optoelectronic sensors such as laser scanners or 3D cameras are often used to detect objects, for example for safety monitoring. Sensors used in safety technology must work particularly reliably and therefore meet high safety requirements, for example the EN13849 standard for machine safety and the device standard IEC61496 or EN61496 for non-contact protective devices (ESP). To meet these safety standards, a number of measures must be taken, such as secure electronic evaluation through redundant, diverse electronics, function monitoring or special monitoring of contamination of optical components.

In the DE 10 2007 007 576 A1 A machine is secured by a large number of laser scanners taking a three-dimensional image of its workspace and comparing this actual state with a target state. The laser scanners are positioned on tripods at different heights at the edge of the work area. 3D cameras can also be used instead of laser scanners.

From the DE 198 43 602 A1 a method and a device for detecting the movements of process units during a production process in a predetermined evaluation area are known. At least two cameras installed in a fixed location in the evaluation area are used. Location coordinates of each process unit are continuously recorded and a translation vector describing the movement of the respective process unit is determined for each location coordinate.

The US 9,804,576 B2 discloses a detection-based, industrial automation control that is set up to detect people's movements, derive them for the future and compare them with planned automation commands in order to derive further safety-related actions (alarms or changed control commands), if necessary, using 3D cameras to detect the people's movements.

The DE 10 2006 048 163 B4 describes a camera-based monitoring of moving machines and/or moving machine elements to prevent collisions, using an image capture system to capture image data of the machine and/or the moving machine elements. The image capture system can in particular be a multiocular camera system; LiDAR, RADAR or ultrasonic sensors are also mentioned as possible image capture systems.

Another area of application for object detection using optoelectronic sensors is traffic monitoring, for example for checking vehicle regulations, calculating tolls, traffic management or statistical traffic monitoring. It is known to detect vehicles in flowing traffic on so-called measuring bridges, which have, for example, optical sensors such as laser scanners or cameras in order to record the vehicle contour. In a corresponding image evaluation, the number and position of the vehicle axles can then be determined, for example. The EP3183721 B1 discloses a system for contactless axle counting of a vehicle using optical methods. The system is camera-based and uses complex image processing, although additional sensors can be used to relieve the burden on image processing in terms of detecting vehicles. However, the known visual methods are not particularly robust against bad weather conditions. Since 3D methods generally work with a lower resolution, they are particularly sensitive to snow and rain because the detection of snow and splash water significantly distorts the measurement data. Overall, the accuracy of axis detection can be unsatisfactory and the computational effort for image processing is usually high.

The image capture systems commonly used in the prior art for object capture, in particular laser scanners and camera systems, also have disadvantages, which will be discussed in more detail below.

Laser scanners or LiDAR (Light Detection And Ranging) sensors are usually based on direct light transit time measurement. A light pulse is emitted by the sensor, reflected on an object and detected again by the sensor. The transit time of the light pulse is determined by the sensor and the distance between the sensor and the object is estimated using the speed of light in the propagation medium (usually air). Since the phase of the electromagnetic wave is not taken into account, this is referred to as an incoherent measuring principle. In the case of an incoherent measurement, there is a need to build up pulses from many photons in order to receive the reflected pulse with a sufficient signal-to-noise ratio. In an industrial environment, the number of photons within a pulse is usually limited by eye protection. As a result, there are trade-offs between maximum range, minimum remission of the object, integration time and the requirements for the signal-to-noise ratio of the sensor system. Incoherent radiation at the same wavelength (ambient light) also has a direct effect on the dynamic range of the light receiver. Examples of incoherent radiation at the same wavelength are the sun, similar sensor systems, or the identical sensor system via multipath propagation, i.e. unwanted reflections.

Camera systems known from the prior art are based on measurement principles such as stereoscopy or indirect light transit time measurement. With indirect time-of-flight measurement, the phase difference of an AMCW (Amplitude Modulated Continuous Wave) transmission signal and its time-delayed copy are determined after reflection with an object. The phase difference corresponds to the transit time of light and can be converted into a distance value using the speed of light in the propagation medium. Both stereoscopy and indirect time-of-flight measurement are also incoherent measurement methods with the disadvantages mentioned above.

Millimeter wavelength radar sensors are based on a frequency-modulated continuous wave measurement principle (FMCW) and can also determine the radial velocities of a detected object using the Doppler effect. The biggest disadvantage of millimeter wavelength radar sensors compared to optical technologies is the significantly longer wavelength and therefore lower spatial resolution. In addition, regulatory regulations limit the radial resolution by limiting the bandwidth and, in a MIMO (Multiple Input Multiple Output) radar system, the number of available virtual antennas (product of the number of transmitting and receiving antennas) the angular resolution.

It is therefore the object of the invention to improve a device for detecting objects.

This object is achieved by a device according to claim 1.

The device according to the invention for detecting objects in a surveillance area has at least one optoelectronic sensor, which is designed as a frequency modulated continuous wave (FMCW) LiDAR sensor. The basics of FMCW-LiDAR technology can be found, for example, in the scientific publication “Linear FMCW Laser Radar for Precision Range and Vector Velocity Measurements” (Pierrottet, D., Amzajerdian, F., Petway, L., Barnes, B., Lockard, G. , & Rubio, M. (2008). Linear FMCW Laser Radar for Precision Range and Vector Velocity Measurements. MRS Proceedings, 1076, 1076-K04-06. doi:10.1557/PROC-1076-K04-06) or the doctoral thesis “Realization of Integrated Coherent LiDAR” (T. Kim, University of California, Berkeley, 2019. https://escholarship.org/uc/item/1d67v62p).

In contrast to a LiDAR sensor based on the time-of-flight measurement of laser pulses, an FMCW LiDAR sensor does not emit pulsed, but continuous light beams into the monitoring area, which during a measurement, i.e. a discrete-time scanning of a measuring point in the monitoring area, produce a predetermined frequency modulation, i.e a temporal change in the wavelength of the transmitted light. The measurement frequency is typically in the range from 10 to 30 Hz. The frequency modulation can, for example, be designed as periodic up and down modulation. Transmitted light reflected from measuring points in the monitoring area has a time offset compared to the emitted transmitted light corresponding to the light transit time, which depends on the distance of the measuring point from the sensor and is accompanied by a frequency shift due to the frequency modulation. In the FMCW LiDAR sensor, emitted and reflected light is coherently superimposed, whereby the distance of the measuring point from the sensor can be determined from the superimposed signal. Compared to pulsed or amplitude-modulated incoherent LiDAR measuring principles, the measuring principle of coherent overlay has, among other things, the advantage of increased immunity to extraneous light from, for example, other optical sensors/sensor systems or the sun. Compared to radar sensors with wavelengths in the millimeter range, the spatial resolution is improved, making the geometric properties of an object measurable.

If a measuring point moves towards or away from the sensor at a radial speed, the reflected transmitted light also has a Doppler shift. An FMCW LiDAR sensor can determine this change in the transmitted light frequency and use it to determine the distance and the radial speed of a measuring point in a single measurement, i.e. a one-time scanning of a measuring point, while a LiDAR sensor based on the time-of-flight measurement of laser pulses can determine the Radial velocity at least two measurements, i.e. two time-spaced scans of the same measuring point, are necessary.

When scanning a three-dimensional monitoring area in discrete time and space, an FMCW LiDAR sensor can record the following measurement data: $$M_{j,k,l}=\left(\begin{array}{l}{r}_{j,k,l}\\ v_{j,k,l}^r\\ I_{j,k,l}\end{array}\right).$$

Here r _j,k,l denote the radial distance, v ^r _j,k,l the radial velocity and I _j,k,l the intensity of each spatially discrete measuring point j, k with a two-dimensional position specified by the azimuth angle φ and polar angle θ (φ _j , θ _k ) for each discrete-time sampling I. For ease of reading, the index n is used below for a one-time discrete-time sampling of a spatially discrete, two-dimensional measuring point (φ _j, θ _k ) in the three-dimensional monitoring area.

To evaluate the measurement data recorded by the FMCW LiDAR sensor, the device according to the invention has a control and evaluation unit which is designed to segment the measurement points and assign them to at least one object and at least one object segment associated with the object. Object segments are, for example, individual parts of an object comprising several parts, for example the torso and limbs of a human body, components of a robot arm or wheels of a vehicle.

The control and evaluation unit is further designed to determine a movement pattern of the first object segment using spatially resolved radial velocities of measuring points assigned to at least a first object segment. A movement pattern is to be understood as a characteristic movement of the object segment, for example a rotation or a relative movement of the object segment to the object itself or to another object segment, which results from a radial velocity profile of the measuring points assigned to the object segment. A corresponding characteristic self-motion or characteristic radial velocity profile can, for example, be stored as a predetermined movement pattern in the control and evaluation unit, recorded in a teaching process or determined during operation of the device using methods of machine learning or artificial intelligence.

The use of the spatially resolved radial velocities to determine a movement pattern of the measuring points assigned to a first object segment has the advantage that the movement pattern of the object segment can be determined quickly and reliably.

The determination of a movement pattern of an object segment, in particular the determination of a relative movement between different object segments or from object segments to the object itself, can be used, for example, in a robot to determine future robot poses.

As part of traffic monitoring, for example, the movement pattern of a vehicle's wheels can be used to determine whether the vehicle's wheels are turning or not. For example, axle detection can be improved because the wheels of a raised axle do not rotate or do not rotate at the same speed as wheels that are in contact with the road.

The segmentation and assignment of the measuring points to at least one object and/or at least one object segment associated with the object can preferably be carried out using the spatially resolved radial velocity of the measuring points. By using the spatially resolved radial velocity as an additional parameter, improved segmentation of the measurement data is possible.

The control and evaluation unit can further be designed to determine radial velocities of the at least one object and/or the at least one object segment associated with the object and to extract features of the object and/or the object segment that are based on the radial velocities of the object or the object segment . The extracted features can be, for example, statistical measures such as mean or standard deviation, higher moments, or histograms of the radial velocities of the object and/or object segment, which can be characteristic of an object and/or object segment movement.

Advantageously, the control and evaluation unit can be designed to use the features based on the radial velocities of the object and/or the object segment to classify the object and/or the object segment. These additional features enable an improved classification of the object and/or the object segment.

In one embodiment, the control and evaluation unit can be designed to filter the measurement data using the radial velocities of the measurement points. Thus can The computing effort can be reduced through data reduction even before the measuring points are segmented. Filtering can be carried out, for example, by discarding measuring points with a radial velocity smaller, larger or equal to a predetermined threshold value and not subjected to any further evaluation. For example, in the case of an anti-collision function, objects and/or object segments that move with the sensor (v ^r = 0) or move away from the sensor (v ^r > 0) can be discarded.

The FMCW LiDAR sensor can be arranged stationary and scan a predetermined monitoring area. Preferably, at least one further FMCW LiDAR sensor can be provided, which scans a further monitoring area, whereby the monitoring areas can overlap. This makes it possible to avoid shadows or blind spots in which object detection is not possible. If two or more FMCW LiDAR sensors are arranged relative to one another in such a way that measurement beams that are orthogonal to one another can be generated, a velocity vector of an object scanned by these measurement beams can be determined by calculating these pairs of measurement beams in the plane spanned by the measurement beams that are orthogonal to one another.

The FMCW LiDAR sensor can be arranged on a machine, in particular on an automated guided vehicle (AGV) or on a robot. The robot can be in motion as a whole (mobile robot) or can carry out movements using different axes and joints. The sensor can then follow the machine's movements and scan a changing monitoring area.

The sensor can be designed as a safety sensor in the sense of the standards mentioned in the introduction or comparable standards. The control and evaluation unit can be integrated into the sensor or connected to it, for example in the form of a safety controller or a higher-level controller that also communicates with the machine controller. At least parts of the functionality can also be implemented in a remote system or a cloud.

The sensor can preferably be attached to or near a dangerous machine part, such as a tool tip. For example, if it is a robot with many axes, their interaction is not relevant to the sensor, as the sensor simply tracks the resulting movement at the location of the danger.

In a further development of the invention, several optoelectronic sensors can be attached to the machine in order to determine the movement of moving parts of the machine. This means that even complex machines can be monitored where a specific determination of movement is not sufficient. An example is a robot with multiple robotic arms and possibly joints. At least one stationary optoelectronic sensor, i.e. one that does not move with the machine, can also monitor the machine.

In a further embodiment of the invention, the device can be set up to monitor a roadway. The device can have one or more FMCW LiDAR sensors, which are arranged above and/or next to a lane of a roadway to be monitored so that one or both sides of a vehicle can be detected when the vehicle uses the FMCW LiDAR -Sensors happened. The control and evaluation device can in particular be set up to determine a speed of the vehicle and/or a movement pattern, in particular a rotation of the vehicle's wheels.

• 1 ein Beispiel für eine Radialgeschwindigkeitsmessung mit einem FMCW-LiDAR-Sensor;
• 2 eine schematische Darstellung einer erfindungsgemäßen Vorrichtung zur Überwachung eines Roboters;
• 3 eine schematische Darstellung einer erfindungsgemäßen Vorrichtung zur Verkehrsüberwachung;
• 4 ein Ablaufdiagram für eine beispielhafte erfindungsgemäße Verarbeitung von Messdaten eines FMCW-LiDAR-Sensors;
• 5 ein beispielhaftes Ablaufdiagramm zur Überwachung einer Bewegung eines Roboters;

The invention will be explained in more detail below with regard to further features and advantages using exemplary embodiments and with reference to the accompanying drawing. The illustrations in the drawing show:

• 1 an example of a radial velocity measurement with an FMCW LiDAR sensor;
• 2 a schematic representation of a device according to the invention for monitoring a robot;
• 3 a schematic representation of a traffic monitoring device according to the invention;
• 4 a flowchart for an exemplary processing according to the invention of measurement data from an FMCW LiDAR sensor;
• 5 an exemplary flowchart for monitoring movement of a robot;

In 1 the concept of radial velocity measurement with an FMWC LIDAR sensor 12 is shown using a three-dimensional example. If an object 38 moves along a direction of movement 40 relative to the FMCW LiDAR sensor 12, the FMCW LiDAR sensor 12 can, in addition to the radial distance r and the intensity 1, be scanned once in a time-discrete manner with a transmitting light beam 14 at an azimuth angle φ and a polar angle θ Measuring point 20 determines the radial velocity v ^r of the measuring point 20 of the object 38 in the direction of the FMCW LiDAR sensor 12. This Information is available directly with a measurement, i.e. a discrete-time sampling of the measuring point 20. In order to identify moving objects, in contrast to measurement methods that only provide spatially resolved radial distances, i.e. three-dimensional positions, there is no need for a second measurement and in particular the need to first determine the measurement points in the measurement data of the second measurement that correspond to the measurement points of the first measurement are equivalent to.

In the case of a static FMCW LiDAR sensor, each measuring point with a radial velocity of zero is usually assigned to a static object, as long as it does not move tangentially to the sensor's measuring beam. Due to the finite object extent and the high spatial resolution of the FMCW LiDAR sensor, practically every moving object will have at least one measuring point 20 with a non-zero radial velocity v ^r _n to the FMCW LiDAR sensor 12. Therefore, a measurement from the FMCW LiDAR sensor 12 can be used to distinguish between static and moving objects, or in mobile applications, objects that are moving away or approaching. For example, in anti-collision monitoring, measuring points or objects that move away can be discarded. Corresponding data reduction reduces the computational effort required for further evaluation of the measurement data.

2 shows a schematic representation of a device 10 according to the invention for monitoring a robot 24. An FMCW LiDAR sensor 12 sends out transmitted light beams 14.1, ..., 14.n into a three-dimensional monitoring area 16 and generates measurement data M _n 18 from measuring points 20.1, . .., 20.n reflected or remitted transmitted light in the monitoring area 16 back to the FMCW LiDAR sensor 12. For illustration purposes, a limited number of exemplary transmitted light beams 14.1, ..., 14.n and measuring points 20.1, ..., 20.n are shown; the actual number results from the size of the monitoring area 16 and the spatial resolution of the scanning. The measuring points 20.1, ..., 20.n can represent people 22, robots 24, or boundaries of the monitoring area such as floors 30 or walls in the monitoring area 16.

The measurement data M _n 18 of the FMCW LiDAR sensor 12 received by the control and evaluation unit 32 includes, for each discrete-time scanning, in addition to the radial distances r _n and the intensities I _n, i.e. the remitted or reflected amount of transmitted light, in particular the radial velocities v ^r _n der Measuring points 20.1, ..., 20.n, whereby radial velocity v ^r _n denotes the speed component of a measuring point 20.1, ..., 20.n, with which the measuring point 20.1, ..., 20.n relates to the FMCW -LiDAR sensor 12 moved to or away from FMCW LiDAR sensor 12.

The measurement data M _n 18 are evaluated by a control and evaluation unit 32, the control and evaluation unit 32 being set up to determine movement patterns of the object segments 22.1, 22.2 using the radial speeds v ^r _n of the measuring points 20.1, ..., 20.n , 22.3, 24.1, 24.2, 24.3 to detect, for example a pivoting movement 26 of the robot arm 24.3. Based on the detection, the control and evaluation unit 32 can generate a safety-related signal to trigger a safety-related action. The safety-related action can be, for example, the activation of a warning light 34 or the stopping of the robot 24. In the exemplary embodiment, the control and evaluation unit 32 is connected directly to the warning light 34 and the robot 24, so it triggers the safety-related action itself. Alternatively, the control and evaluation unit 32 can pass on a safety-related signal to a higher-level safety controller (not shown) via an interface 36, or the control and evaluation unit 32 can itself be part of a safety controller.

3 shows a schematic representation ((a) top view, (b) side view) of a device 90 according to the invention for traffic monitoring. One or more FMCW lidar sensors 92a, 92b are arranged on a so-called toll or traffic sign bridge 94 in order to detect a vehicle, in this case a truck 96, on a roadway 98. The FMCW LiDAR sensors 92a, 92b are arranged above or to the side of the roadway 98 to be monitored so that one or both vehicle sides 96.1, 96.2, as well as wheels 96.3, 96.4, 96.5 of the vehicle 96 are detected when the vehicle 96 enters the monitoring area 100a , 100b of the sensors 92a, 92b happens.

The FMCW LiDAR sensors 92a, 92b send out transmitted light beams 102.1, ..., 102.n into the three-dimensional monitoring areas 100a, 100b of the sensors 92a, 92b and generate measurement data from measuring points 104.1, ..., 104.n in the Monitoring areas 100a, 100b back to the sensors 92a, 92b reflected or remitted transmitted light. The measurement data is evaluated by a control and evaluation unit (not shown).

Measuring points 104.1, 104.2 on static object segments of the vehicle 96, for example the vehicle side 96.1, have radial velocities in the measurement data that differ from the speed of the vehicle in the direction of movement as in 1 described, solely based on the lateral angle difference between the direction of movement and the direction of measurement. Measuring points 104.4, 104.5 on object segments with an additional movement relative to the vehicle, such as rotating wheels 96.3, 96.4, have radial velocities in the measurement data that differ significantly from the radial velocities of the static object segments of the vehicle. Static object segments of the vehicle can in particular include wheels 96.5 on a raised axle of the vehicle 96, since these do not rotate.

A look at the side view of the vehicle 96 shown in b) makes this clear. A measuring point 104.4 at the highest point of the wheel 96.4 in contact with the road has, in addition to the movement speed v ₀ of the vehicle, a rotation component v _r , which results in a total speed of v ₀ ⁺ v _r . At the measuring point 104.5, at which the wheel 96.4 touches the road 98, the rotation speed v _r and the movement speed v ₀ of the vehicle cancel each other out due to the direction of rotation. The radial speed of this measuring point 104.5 on the rotating wheel 96.4 is therefore measured by the sensor as 0. This results in a characteristic movement pattern or radial speed profile for a rotating wheel, which can be determined from the measured radial speeds.

A wheel 96.5 on a raised axle will generally not rotate and therefore has the same speed v ₀ at all measuring points 104.3, 104.n, which corresponds to the movement speed v ₀ of the vehicle 96.

This means that rotating wheels or axles on the vehicle can be reliably detected, which enables improved axle counting.

4 shows in a flow chart 42 an exemplary processing according to the invention of the measurement data recorded by the FMCW LiDAR sensor by the control and evaluation unit. After receiving 44 the measurement data, the measuring points 20.1, ..., 20.n, 104.1, ..., 104.n are segmented in a segmentation step 46 and formed into objects 22, 24, 96 and/or object segments 22.1, 22.2, 22.3 , 24.1, 24.2, 24.3, 96.1, ... 96.5 summarized, in addition to the location coordinates and intensities of the measuring points usually used for segmentation 46, in particular the spatially resolved radial velocities v ^r _n of the measuring points 20.1, ..., 20.n 104.1, . .., 104.n can be taken into account. Object segments can, for example, be individual moving components 24.1, 24.2, 24.3 of a robot 24, body parts 22.1, 22.2, 22.3 of a person 22 or wheels 963, 96.4, 96.5 of a vehicle 96.

• - Pixelorientierte Verfahren im Grauwertbild mittels Schwellenwert-Verfahren,
• - Kantenorientierte Verfahren wie dem Sobel- oder Laplace-Operator und einer Gradientensuche,
• - Regionenorientierte Verfahren wie „Region-Growing“, „Region-Splitting“, „Pyramid Linking“ oder „Split and Merge“,
• - Modellbasierte Verfahren wie bspw. die Hough-Transformation, oder
• - Texturorientierte Verfahren.

The segmentation 46 can be carried out using known methods of digital image processing or machine vision, such as, for example

• - Pixel-oriented methods in the gray value image using threshold value methods,
• - Edge-oriented methods such as the Sobel or Laplace operator and a gradient search,
• - Region-oriented methods such as “region growing”, “region splitting”, “pyramid linking” or “split and merge”,
• - Model-based methods such as the Hough transformation, or
• - Texture-oriented methods.

• - „Fast Range Image-Based Segmentation of Sparse 3D Laser Scans for Online Operation“ (Bogoslavskyi et al., 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems, DOI: 10.1109/IROS.2016.7759050)
• - „Laser-based segment classification using a mixture of bag-of-words“. (Behley et al., 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, DOI: 10.1109/IROS.2013.6696957)
• - „On the segmentation of 3d lidar point clouds“ (Douillard et al., 2011 IEEE International Conference on Robotics and Automation, DOI: 10.1109/ICRA.2011.5979818)

Furthermore, the term “range segmentation” refers to special methods for segmenting three-dimensional data sets. “Range segmentation” is described, for example, in the following scientific publications:

• - “Fast Range Image-Based Segmentation of Sparse 3D Laser Scans for Online Operation” (Bogoslavskyi et al., 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems, DOI: 10.1109/IROS.2016.7759050)
• - “Laser-based segment classification using a mixture of bag-of-words”. (Behley et al., 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, DOI: 10.1109/IROS.2013.6696957)
• - “On the segmentation of 3d lidar point clouds” (Douillard et al., 2011 IEEE International Conference on Robotics and Automation, DOI: 10.1109/ICRA.2011.5979818)

By using the radial velocity v ^r _n in addition to the radial distance r _n and the intensity 1" of the measuring points 20.1, ..., 20.n, 104.1, ..., 104.n, the segmentation 46 of the measuring points 20.1, ... , 20.n, 104.1, ..., 104.n can be done more efficiently and accurately using the methods listed above. For example, measuring points 20.1, ..., 20.n, 104.1, ..., 104.n with radial velocities v ^r _n less than, greater than or equal to a predetermined threshold value are discarded and no further evaluation is carried out. In the case of an anti-collision function, for example, measuring points of an object and / or object segment that move with the sensor (v ^r = 0) or move away from the sensor (v ^r > 0), are discarded.

If an object and/or object segment is scanned through several spatially discrete measuring points and the associated radial velocities can be distinguished, static and dynamic objects and/or object segments can be distinguished and stationary objects and/or object segments such as floors 30, roadways 98 or walls can be identified before or during the segmentation 46 of the measuring points 20.1, ..., 20.n, 104.1, ..., 104.n are discarded and the computing effort is reduced by data reduction.

In the next step, a feature extraction 48 of the objects 22, 24, 30, 96 and/or object segments 22.1, 22.2, 22.3, 24.1, 24.2, 24.3., 96.1, ..., 96.5 defined during the segmentation 46 takes place. Typical features that can be extracted from the objects 22, 24, 30, 96 and/or object segments 22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5 when processing the measurement data are, for example, width, Number of measurement points or length of the perimeter of the objects and/or object segments, or other characteristics, as described, for example, in the scientific publication “A Layered Approach to People Detection in 3D Range Data” (Spinello et al., Proceedings of the Twenty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2010). According to the invention, these features can be expanded to include features that are based on the radial velocities of the objects 22, 24, 30, 96 and/or object segments 22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5. For this purpose, radial velocities of the objects and/or object segments are first determined, for example by applying trigonometric functions to the radial velocities of the measuring points representing the respective object and/or object segment. Statistical measures of the radial velocities of the objects and/or object segments such as mean values, standard deviations, higher moments or histograms, which are characteristic of robot and/or person movements, can then be used as additional object and/or object segment features.

After the feature extraction 48, a classification 50 of the objects 22, 24, 96 and/or object segments 22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5 takes place using known classification methods such as Bayes classifiers, support vector machines or artificial neural networks. Classification involves searching the feature space for groups of features that define an object. The above-listed statistical measures of the radial speed of individual objects 22, 24, 30, 96 and/or object segments 22.1, 22.2, 22.3, 24.1, 24.2, 24.3 96.1, ..., 96.5 can be used in combination with a priori information To define feature spaces that, for example, can classify and thus distinguish people 22 or vehicles 96 based on their radial speed.

In a further step 52, a movement pattern of at least one of the object segments 22.1, 22.2, 22.3, 24.1, 24.2, 24.3 96.1, ..., 96.5 is determined using the radial velocities of the measuring points 20.1, ... assigned to the at least one object segment. 20.n, 104.1, ..., 104.n.

The result of the determination of the movement pattern 52 can be further processed by the control and evaluation unit 32 after the output 54, for example to generate a safety-related signal, to detect the status of an object segment (wheel rotates or not) or via the interface 36 to a higher-level controller (not shown).

5 shows an exemplary flowchart 54 for monitoring a movement of a robot. After receipt 44 of the measurement data M _n, the steps of segmentation 46 of the measurement data M _n , feature extraction 48 and classification 50 take place as described above. For segments 24.1, 24.2, 24.3 of the robot arm identified in the classification 50, a determination 56 of representative variables such as radial distances and intensities takes place and radial velocities of segments 24.1, 24.2, 24.3.

• • Dewan, Ayush, et al. „Rigid scene flow for 3d lidar scans.“ 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2016. oder
• • Liu, Xingyu, Charles R. Qi, and Leonidas J. Guibas. „Flownet3d: Learning scene flow in 3d point clouds.“ Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019.

beschrieben, können die gemessenen Radialgeschwindigkeitswerte direkt zur Erkennung eines Bewegungsmusters verwendet werden, so dass insbesondere zwei zeitlich aufeinanderfolgende Abtastungen M_n,l und M_n,l-1 des Überwachungsbereichs nicht zwingend notwendig sind. In einem Vergleichsschritt 60 erfolgt ein Vergleich des Bewegungsmusters 58 mit a priori Informationen über erwartete Soll-Bewegungen der Segmente 24.1, 24.2., 24.3 des Roboterarms. Bei negativem Ergebnis des Vergleichs 60 (beispielsweise Bewegungsabweichung über ein vorgegebenes Toleranzmaß) wird eine sicherheitsgerichtete Aktion 62, beispielsweise ein Abschalten des Roboters 24 eingeleitet.Based on the measured radial velocities of the measuring points assigned to the previously classified segments 24.1, 24.2, 24.3, a movement pattern 58 is recognized. In contrast to the typical determination of a “rigid scene flow” based on 3D position data such as in

• • Dewan, Ayush, et al. “Rigid scene flow for 3d lidar scans.” 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2016. or
• • Liu, Xingyu, Charles R. Qi, and Leonidas J. Guibas. “Flownet3d: Learning scene flow in 3d point clouds.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019.

described, the measured radial velocity values can be used directly to detect a movement pattern, so that in particular two consecutive scans M _n,l and M _n,l-1 of the monitoring area are not absolutely necessary. In a comparison step 60, the movement pattern 58 is compared with a priori information about expected target movements of the segments 24.1, 24.2, 24.3 of the Robot arm. If the result of the comparison 60 is negative (for example movement deviation beyond a predetermined tolerance level), a safety-related action 62, for example switching off the robot 24, is initiated.

QUOTES INCLUDED IN THE DESCRIPTION

This list of documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 102007007576 A1 [0003]
• DE 19843602 A1 [0004]
• US 9804576 B2 [0005]
• DE 102006048163 B4 [0006]
• EP 3183721 B1 [0007]

Claims (9)

Device (10, 90) for detecting at least one object (22, 24, 96) comprising a plurality of object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5) in a monitoring area (16, 100a, 100b) with - at least one FMCW LIDAR sensor (12, 92a, 92b) for emitting transmitted light beams (14, ..., 14.n, 102.1, ..., 102.n) into the monitoring area (16 , 100a, 100b), for scanning a large number of measuring points (20.1, ..., 20n, 104.1, ..., 104.n) and for generating measurement data (18) from the measuring points (20.1, ..., 20n, 104.1, ..., 104.n) remitted or reflected transmitted light, the measurement data (18) being radial velocities (v ^r _n ) of the measuring points (20.1, ..., 20n, 104.1, ..., 104.n) include, - a control and evaluation unit (32) for evaluating the measurement data (18), the control and evaluation unit (32) being designed to measure the measuring points (20.1, ..., 20n, 104.1, ..., 104 .n) to segment and at least partially assign them to the object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5), characterized in that the control and evaluation unit (32) is further designed to using the radial velocities (v ^r _n ) of the measuring points assigned to the object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5), a movement pattern of at least one of the object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5). Device according to Claim 1 , wherein the control and evaluation unit (32) is designed to segment the measuring points (20.1, ..., 20n, 104.1, ..., 104.n) using the radial velocities (v ^r _n ) and at least partially Object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5). Device according to Claim 1 or 2 , wherein the control and evaluation unit (32) is designed to determine radial velocities of the objects (22, 24, 96) and/or object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5). determine and characteristics of the objects (22, 24, 96) and/or object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5) using the radial velocities of the objects (22, 24, 96) and/or to extract object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5) (48). Device according to Claim 3 , wherein the control and evaluation unit (32) is designed to use the objects (22, 24, 96) and/or object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5). the radial velocities of the objects (22, 24, 96) and/or object segments (22.1, 22.2, 22.3, 24.1, 24.2, 24.3, 96.1, ..., 96.5) to classify (50). Device according to one of the Claims 1 until 4 , wherein the control and evaluation unit (32) is designed to measure the measuring points (20.1, ..., 20n, 104.1, ..., 104.n) using the radial speed (v ^r _n ) of the measuring points (20.1, . .., 20n, 104.1, ..., 104.n). Device according to Claim 5 , wherein the control and evaluation unit (32) is designed to select measuring points (20.1, ..., 20n, 104.1, ..., 104.n) with a radial speed (v ^r _n ) below a predetermined threshold value for the evaluation discard. Device according to one of the preceding claims, wherein the FMCW LiDAR sensor (12) is stationary. Device according to Claim 7 , wherein the device has at least one further FMCW LiDAR sensor with a further monitoring area and the monitoring area (16) at least partially overlaps with the further monitoring area. Device according to one of the preceding claims, wherein the device (90) is set up to detect vehicles (96) on a roadway (98), the control and evaluation device being designed to detect a rotation of wheels (96.3, 96.4, 96.5 ) of the vehicle (96).

Images