DE102018127865A1 - Method for detecting an object using a laser scanner - Google Patents

Abstract

The present invention relates to a method for detecting an object by a laser scanner, with the following method steps: - scanning an environment of the laser scanner by a laser beam, - detecting reflected laser beam light in the form of scanning points (3), - determining at least one suspected object in the surroundings of the laser scanner based on the detected reflected laser beam light in the form of scanning points (3), and - determining a bounding box (1) for the at least one suspected object, characterized by - counting the number of scanning points (3) in the bounding box (1) , and- Validate whether the putative object is a real object or a false positive result by comparing the number of counted sample points (3) with an expected number of sample points (3) for a real object. In this way provides a method to efficiently use an e Validate individual sensor whether a detected suspected object is a real object or a false positive result.

Description

• - Abtasten einer Umgebung des Laserscanners durch einen Laserstrahl,
• - Erfassen von reflektiertem Laserstrahllicht in der Form von Abtastpunkten,
• - Bestimmen mindestens eines mutmaßlichen Objekts in der Umgebung des Laserscanners basierend auf dem erfassten reflektierten Laserstrahllicht in der Form von Abtastpunkten, und
• - Bestimmen eines Begrenzungsrahmens für das mindestens eine mutmaßliche Objekt.

The invention relates to a method for detecting an object by a laser scanner, in particular for autonomous driving, with the following method steps:

• Scanning an environment of the laser scanner by a laser beam,
• Detection of reflected laser beam light in the form of sampling points,
• Determining at least one putative object in the vicinity of the laser scanner based on the detected reflected laser beam light in the form of scanning points, and
• - Determining a bounding box for the at least one suspected object.

The most critical task of autonomous driving is object detection using available sensor data. The detected objects are also used to define the attainable visibility for the sensor used. In laser scanner applications, these objects are used to define the visible space in the field of view (FOV) of the sensor. However, inaccuracies in detection algorithms can also lead to the creation of ghost objects (false positive results), which results in an inaccurate input for the estimation of the visible space.

A false positive result in such a scenario leads to a misinterpretation of the available visible space, which affects the performance of the post-processing steps. The acquisition rate can be improved by using multiple sensors and merging the data available from all sensors. In this regard, is out of the US 9383753 B1 a LIDAR (Light Detection and Ranging) device with a dynamically adjustable angular resolution for use as a sensor is known which provides environmental information for navigation in an autonomous vehicle. A first area of a scan zone is scanned while light pulses are emitted at a first pulse rate, and a second area of the scan zone is scanned while light pulses are emitted at a second pulse rate that is different from the first pulse rate. Information is received from the LIDAR device indicating the time delays between the transmission of the light pulses and the reception of the corresponding returning light pulses. A three-dimensional point map is generated, the resolution of the point map being based on the first pulse rate in the first area and on the second pulse rate in the second area.

Similarly, the US 9097800 B1 a LIDAR (Light Detection and Ranging) device which is assigned to an autonomous vehicle and scans a scanning zone by emitting light pulses and receiving reflected signals corresponding to the light pulses. The reflected signals indicate a three-dimensional point map of the distribution of reflecting points in the scan zone. A RADAR (Radio Detection and Ranging) device scans an area of the scan zone that corresponds to a reflective feature indicated by the three-dimensional dot map. Solid objects are distinguished from non-massive reflective features based on a reflected radio signal that corresponds to the reflective feature. Positions of features indicated by the reflected radio signals are projected onto estimated positions during scanning by the LIDAR device in accordance with a relative movement of the radio reflecting features indicated by a frequency shift in the reflected radio signals.

However, the methods mentioned above not only increase the cost of hardware, but also introduce interdependencies between the sensors used.

The invention is therefore based on the object of specifying a method which, using a single sensor, efficiently validates whether a detected suspected object is a real object or a false positive result.

This object is achieved by the subject matter of the independent claims. Preferred embodiments are described in the subclaims.

• - Abtasten einer Umgebung des Laserscanners durch einen Laserstrahl,
• - Erfassen von reflektiertem Laserstrahllicht in der Form von Abtastpunkten,
• - Bestimmen mindestens eines mutmaßlichen Objekts in der Umgebung des Laserscanners basierend auf dem erfassten reflektierten Laserstrahllicht in der Form von Abtastpunkten, und
• - Bestimmen eines Begrenzungsrahmens für das mindestens eine mutmaßliche Objekt, gekennzeichnet durch
• - Zählen der Anzahl von Abtastpunkten im Begrenzungsrahmen, und
• - Validieren, ob es sich bei dem mutmaßlichen Objekt um ein reales Objekt oder ein falsch positives Ergebnis handelt, durch Vergleichen der Anzahl gezählter Abtastpunkte mit einer erwarteten Anzahl von Abtastpunkten für ein reales Objekt.

According to the invention, a method for detecting an object by a laser scanner is therefore specified, which has the following method steps:

• Scanning an environment of the laser scanner by a laser beam,
• Detection of reflected laser beam light in the form of sampling points,
• Determining at least one putative object in the vicinity of the laser scanner based on the detected reflected laser beam light in the form of scanning points, and
• - Determining a bounding box for the at least one putative object, characterized by
• - counting the number of sample points in the bounding box, and
• - Validate whether the suspected object is a real object or a false positive by comparing the number of counted sample points with an expected number of sample points for a real object.

With algorithms, it is not always possible to achieve a 100% acquisition rate due to numerous unavoidable factors that affect performance. For example, obscurations, skipped picture frames, etc. can significantly affect the acquisition rate by generating false positive and false negative results. The present invention is therefore concerned with reducing false positive results by validating the detected objects by counting the number of sample points associated with the bounding box of the object and performing a comparison with expected data for the sample points without incurring additional costs and interdependencies. that can occur when multiple sensors are used.

In general, various types of laser scanners can be used with the invention. A laser scanner is preferably a LIDAR sensor. LIDAR sensors have proven to be efficient and versatile for automotive applications and are therefore well suited for the present invention.

• - Bereitstellen einer Nachschlagetabelle für die erwartete Anzahl von Abtastpunkten für reale Objekte verschiedener Klassifizierungen, und
• - Vergleichen der Anzahl gezählter Abtastpunkte mit der erwarteten Anzahl von Abtastpunkten für die realen Objekte mit verschiedenen Klassifizierungen. Auf diese Weise wird das Validieren, ob es sich bei dem mutmaßlichen Objekt um ein reales Objekt oder um ein falsch positives Ergebnis handelt, auf der Basis verschiedener Klassen erwarteter Objekte ausgeführt, die in der Nachschlagetabelle definiert sind. Die verschiedenen Klassen erwarteter Objekte, die durch die Nachschlagetabelle bereitgestellt werden, werden im Voraus auf der Basis der gemessenen Anzahl von Abtastpunkten für beispielhafte reale Objekte bestimmt. Auf diese Weise wird die Validierung weniger fehleranfällig und effizienter. In dieser Hinsicht ist die erwartete Anzahl von Abtastpunkten vorzugsweise durch einen Bereich von Anzahlen von Abtastpunkten mit einem oberen und einem unteren Grenzwert gegeben. Daher kann ein statistisches Verfahren in Bezug auf die erwartete Anzahl von Abtastpunkten für Objekte verschiedener Klassen verwendet werden.

The comparison of the number of counted sampling points with an expected number of sampling points for a real object can be done in different ways. According to a preferred embodiment of the invention, however, the method has the following further method steps:

• Providing a look-up table for the expected number of sampling points for real objects of different classifications, and
• - Compare the number of counted sampling points with the expected number of sampling points for the real objects with different classifications. In this way, validation as to whether the suspected object is a real object or a false positive result is performed based on various classes of expected objects defined in the lookup table. The various classes of expected objects provided by the lookup table are determined in advance based on the measured number of sampling points for exemplary real objects. This makes validation less error prone and more efficient. In this regard, the expected number of sampling points is preferably given by a range of numbers of sampling points with an upper and a lower limit. Therefore, a statistical method regarding the expected number of sampling points for objects of different classes can be used.

Furthermore, the method according to the invention not only enables a highly reliable validation of a putative object. According to a preferred embodiment of the invention, the method further comprises the step of determining the classification of the putative object that has been validated based on the lookup table as a real object. This also makes it clear which classification the validated object belongs to. The classification is preferably at least one classification among pedestrians, motorcycles, cars and trucks.

• - Erfassen des reflektierten Laserstrahllichts in Form von Abtastpunkten in mehreren Schichten, die mit verschiedenen Entfernungszonen vom Laserscanner in Beziehung stehen,
• - Bestimmen, in welchen Schichten sich der Begrenzungsrahmen befindet, und
• - Zählen der Abtastpunkte nur in den Schichten, in denen sich der Begrenzungsrahmen befindet. Auf diese Weise wird kein Arbeitsaufwand darauf aufgewendet, Abtastpunkte in anderen Schichten zu zählen, als in denjenigen, in denen sich der Begrenzungsrahmen tatsächlich befindet. Dies spart Rechenzeit und macht das Verfahren somit schneller und effizienter. Vorzugsweise erfolgt das Bestimmen, in welchen Schichten sich der Begrenzungsrahmen befindet, dadurch, dass bestimmt wird, wo sich der nächstgelegene Punkt des Begrenzungsrahmens und der am weitesten entfernte Punkt des Begrenzungsrahmens in Bezug auf den Laserscanner befinden.

The above-mentioned object of the invention can already be achieved by the method explained above. According to a preferred embodiment of the invention, however, the method also has the following method steps:

• Detection of the reflected laser beam light in the form of scanning points in several layers which are related to different distance zones from the laser scanner,
• - determine which layers the bounding box is in, and
• - Counting the sampling points only in the layers in which the bounding box is located. In this way, no work is required to count sample points in layers other than those in which the bounding box is actually located. This saves computing time and makes the process faster and more efficient. Preferably, determining which layers the bounding box is in is done by determining where the closest point of the bounding box and the most distant point of the bounding box are with respect to the laser scanner.

• - Bestimmen des Winkelbereichs des mutmaßlichen Objekts relativ zum Laserscanner durch Erfassen des am weitesten rechts liegenden Punktes und des am weitesten links liegenden Punktes des Begrenzungsrahmens relativ zum Laserscanner. Auf diese Weise wird weitere Information über das validierte Objekt verfügbar, weil auch bestimmt wird, in welchem Winkelsektor sich das validierte Objekt relativ zum Laserscanner befindet.

It is not mandatory for the invention to determine in which angle sector the validated object is located relative to the laser scanner. According to a preferred embodiment of the invention, however, the method has the following step:

• - Determine the angular range of the suspected object relative to the laser scanner by detecting the right most point and the left most point of the bounding box relative to Laser scanner. In this way, further information about the validated object is available because it is also determined in which angle sector the validated object is located relative to the laser scanner.

The invention only requires determining the number of sample points in the bounding box and validating whether the suspected object is a real object or a false positive by comparing the number of counted sample points with an expected number of sample points for a real object . According to a preferred embodiment of the invention, however, not only the number of sampling points in the bounding frame is determined, but also the intensity of the sampling points. In this way, further information for a safe validation of the suspected object can be acquired, i.e. whether it is a real object or a false positive result.

The invention also relates to the use of a method described above in a vehicle that is traveling on a road.

The invention further relates to a non-transitory computer-readable medium which contains instructions stored thereon which, when executed by a processor, implement a method described above.

These and other aspects of the invention will become apparent and explained from the embodiments described below. Individual features set forth in the embodiments may form an aspect of the present invention, alone or in combination. Features of the various embodiments can be transferred from one embodiment to another embodiment.

• 1 schematisch ein Szenario im Sichtfeld eines Laserscanners gemäß einer bevorzugten Ausführungsform der Erfindung;
• 2 schematisch die Erfassung eines Bereichs von Interesse (ROI) gemäß der bevorzugten Ausführungsform der Erfindung; und
• 3 schematisch das Szenario mit unterschiedlichen Schichten im Sichtfeld eines Laserscanners gemäß der bevorzugten Ausführungsform der Erfindung.

Show it:

• 1 schematically a scenario in the field of view of a laser scanner according to a preferred embodiment of the invention;
• 2nd schematically the detection of an area of interest (ROI) according to the preferred embodiment of the invention; and
• 3rd schematically the scenario with different layers in the field of view of a laser scanner according to the preferred embodiment of the invention.

1 shows schematically a scenario in which an object in the field of view 2nd (FOV) of a laser scanner that is mounted on a vehicle, which is referred to below as an ego vehicle. In FOV 2nd is the position of the ego vehicle with the laser scanner at the origin 0 . The object is defined by its bounding box 1 displayed, with the representation of sample points for the object within the bounding box 1 has been omitted. There will be more sample points 3rd recorded that are not related to any objects. The free and thus accessible space that is available for the ego vehicle is shown by a polygon with dashed lines.

A false positive result in such a scenario leads to a misinterpretation of the available visible space, which affects the performance of post-processing steps. Although the acquisition rate can be improved by using multiple sensors and merging the data available from all sensors, it is an advantage of the present invention that only a single sensor, i.e. a laser scanner can be used to avoid additional cost and effort while still validating captured objects can be performed efficiently and reliably.

The algorithm according to the preferred embodiment of the invention for validating object detection uses the bounding box 1 of a detected object, as in 1 is shown. As in 2nd is shown in order to identify sample points that by its bounding box 1 assigned object, an expected area relative to the ego vehicle 4th Are defined. The angle at which the bounding box 1 that defines the object in FOV 2nd can, using the leftmost and rightmost corners of the bounding box 1 of the object can be calculated. This is also in 2nd shown the the angles θ ₁ and θ ₂ represents where the object in FOV 2nd lies. In the present example, the scanning points of the detected object can be between the angles θ ₁ and θ ₂ in the right area of the FOV 2nd lie.

As in 3rd , the laser scanner in accordance with the presently preferred embodiment of the invention described herein detects the reflected laser beam light in the form of sampling points in multiple layers related to different distance zones from the laser scanner located at the origin 0 of in 3rd shown coordinate system is located.

After the angles have been identified between which the bounding box 1 of the object, it is checked which layers are actually on the bounding box 1 of the object. 3rd shows a scenario in which the bounding box 1 of the object in the FOV 2nd of the laser scanner viewed from above (bird's eye view). In the 3rd drawn ellipses put the projection lines 5 represents that for the individual layers L ₁ to L ₇ were calculated that are related to different distance zones from the laser scanner, where L _{1 describes} the layer that is closest to the laser scanner located at the origin 0 located, and L ₇ describes the layer furthest away from the laser scanner.

The captured object is identified by its bounding box 1 whose nearest and farthest corners are represented by ( x _near , y _near ) respectively. (x _far , y _far ) are designated. Out 3rd it can be seen that only layers L ₅ to L ₇ hit the object. Therefore, only these layers L ₅ to L ₇ used to perform the bounding box check, which makes the procedure more effective than running the bounding box check for those layers that do not strike the object and therefore may not contain relevant information for the present method. The bounding box check itself is described below.

The next step involves identifying and counting the sample points from the relevant slices L ₅ to L ₇ that are within the bounding box 1 are located. The main criterion for validating the object, whether it is a real object or a false positive result, according to the present preferred embodiment of the invention, is the number of sample points that are in the bounding box 1 of the object are included.

The information about the class of the object is also available according to the preferred embodiment of the invention. The various classes to which an object can belong according to the preferred embodiment of the invention are “pedestrians, motorcycles, cars and trucks”. Statistical analysis can be performed on the total number of points an object can have depending on its class. A pedestrian will have fewer sample points than a motorcycle, and the expected number of points for a car or truck is even higher. Therefore, according to the preferred embodiment of the invention, a look-up table is provided based on the class of an object and the expected number of sample points it may have.

According to the preferred embodiment of the invention, this information is also used to verify whether the detected object is a real object or a false positive result. This means that a detected object can only be verified as a real object if its number of sample points is related to one of the ranges of the number of sample points in the lookup table. Therefore, there are a number of sample points that can occur that are not related to any of these ranges of sample points, which means that the bounding box with these sample points is not related to a real object, but rather to a false positive result would be verified. In this way, according to the preferred embodiment of the invention, a method for efficiently validating whether a detected putative object is a real object or a false positive result is provided, which can be implemented using a single sensor. Additional costs due to multiple sensors can thus be avoided and there can be no interdependencies between multiple sensors.

As also mentioned above, although the invention only determines the number of sample points in the bounding frame 1 and validating the putative object, whether it is a real object or a false positive result, by comparing the number of counted sample points with an expected number of sample points for a real object, according to another preferred embodiment of the invention, not only that Number of sample points in the bounding box 1 determined, but also the intensity of the sampling points. This provides further information for reliably validating the detected putative object, whether it is a real object or a false positive result.

Reference symbol list

1

Bounding box

2nd

Field of view

3rd

Sampling points

4th

Ego vehicle

5

Projection lines

L ₁ - L ₇

layers

θ ₁ , θ ₂

Angular range

(X _near , y _near )

closest point of the bounding box

(x _far , y _far )

furthest point of the bounding box

QUOTES INCLUDE IN THE DESCRIPTION

This list of documents listed by the applicant has been generated automatically and is only included 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.

Patent literature cited

• US 9383753 B1 [0003]
• US 9097800 B1 [0004]

Claims (10)

Method for detecting an object by a laser scanner, comprising the following method steps: - scanning an environment of the laser scanner by a laser beam; - Detection of reflected laser beam light in the form of sampling points (3); - Determining at least one suspected object in the vicinity of the laser scanner based on the detected reflected laser beam light in the form of scanning points (3); and - determining a bounding box (1) for the at least one suspected object, characterized by - counting the number of sampling points (3) in the bounding box (1); and - validating whether the putative object is a real object or a false positive result by comparing the number of counted sampling points (3) with an expected number of sampling points (3) for a real object. Procedure according to Claim 1 , the expected number of sampling points (3) being given by a range for the number of sampling points (3). Procedure according to Claim 2 , with the following method steps: - providing a look-up table for the expected number of sampling points (3) for real objects of different classifications; and comparing the number of counted sampling points (3) with the expected number of sampling points (3) for the real objects with different classifications. Procedure according to Claim 3 with the following method step: - determining a classification of the putative object that has been validated as a real object, based on the lookup table. Procedure according to Claim 4 , whereby the classification is at least one of the following classifications: "Pedestrian, motorcycle, car and truck". Method according to one of the preceding claims, with the following method steps: - detecting the reflected laser beam light in the form of scanning points (3) in a plurality of layers (L ₁ - L ₇ ) which are related to different distance zones from the laser scanner; - Determine in which layers (L ₁ - L ₇ ) the bounding frame (1) is located; and - counting the sampling points (3) only in the layers (L ₁ - L ₇ ) in which the bounding frame (1) is located. Procedure according to Claim 6 , with the following method step: - Determine in which layers (L ₁ - L ₇ ) the bounding box (1) is located by determining where the closest point (x _near , y _near ) of the bounding box (1) and where the am farthest point (x _far , y _far ) of the boundary box (1) are arranged in relation to the laser scanner. A method according to any one of the preceding claims, including the step of: determining the angular range (θ ₁ , θ ₂ ) of the suspected object relative to the laser scanner by detecting the right most point and the left most point of the bounding box (1) relative to the laser scanner. Use of a method according to one of the preceding claims in a vehicle traveling on a road. Non-transitory computer readable medium comprising instructions stored thereon which, when executed by a processor, are a method according to one of the Claims 1 to 8th to implement.

Images