A multi-eye 3D laser radar imaging detection method and device

Abstract

The present invention discloses a multi-eye 3D laser radar imaging detection method and device, including: emitting laser to a target, controlling a receiving system to receive the emitted light reflected by the target, and forming images with different depths of field; generating point cloud data according to the images with different depths of field, and extracting features from the point cloud data based on a convolutional neural network to generate a high-resolution three-dimensional image; and performing target detection on the high-resolution three-dimensional image based on an LDA topic model to generate a high-resolution three-dimensional image of the target. The present invention adopts multi-CMOS three-dimensional imaging technology, and multiple eyes perform data detection simultaneously, which reduces the data collection time, improves the timeliness of the receiving stage, and provides a basis for high-speed target clear imaging.

Description

Technical Field

[0001] The present invention relates to the field of laser radar imaging detection technology, and in particular to a multi-eye 3D laser radar imaging detection method and device. Background Art

[0002] Imaging LiDARs currently use mostly mechanical scanning or flash / snapshot methods. Mechanical scanning (linear / point-by-point scanning) exchanges time for high spatial resolution. Flash / snapshot methods are currently divided into direct detection and indirect detection. Direct detection is limited by the size of the device, and the current mature device index is 64*64. This method still has a certain gap for high-resolution imaging. For indirect detection, it is usually based on CCD sensors. Although it can improve resolution, it is usually modulated imaging, which requires multi-frame imaging and signal solution, and its reliability and timeliness are difficult to guarantee.

[0003] In order to solve the reliability and timeliness issues of flash / snapshot imaging lidar, in actual use, two or more lidar imaging radars are usually installed on the target application. The final high-resolution laser image is formed through equipment redundancy, target multi-device imaging comparison, etc. However, this method increases the investment in the application and lacks economy.

[0004] At the same time, in traditional imaging lidar, dozens of slice images are generally required to obtain a three-dimensional target image, which is not conducive to high-speed target detection. At the same time, the processing of dozens of slice images requires a lot of computing power, which increases the cost of the device.

[0005] In addition, in the traditional image recognition method of imaging laser radar, the sample image is often extracted from the shape, aspect ratio or area of ​​the connected area, and then these features are input into the neural network to train the network. After the test image is processed in the same way, the trained network is used to judge the test image. In order not to lose the detailed features of the image, the process of extracting the preset features is sometimes ignored, and all the pixels in the image are directly input into the network. However, the former of the above methods cannot guarantee the extraction of effective and important features, and the latter is too cumbersome and brings a large amount of redundant information. Summary of the invention

[0006] In order to overcome the problems of high precision, reliability and timeliness of imaging laser radar in the prior art, the present invention provides a multi-eye 3D laser radar imaging detection method and device, which can more accurately and quickly obtain the relative speed and relative distance of the detected target.

[0007] The present invention discloses a multi-eye 3D laser radar imaging detection method, comprising:

[0008] Step 1: Launch laser at the target;

[0009] Step 2, receiving the emitted light reflected by the target to form images with different depths of field;

[0010] Step 3: Generate point cloud data based on images with different depths of field, and extract features from the point cloud data based on a convolutional neural network to generate a high-resolution three-dimensional image;

[0011] Step 4: Perform target detection on the high-resolution three-dimensional image based on the LDA topic model to generate a high-resolution three-dimensional image of the target.

[0012] As a further improvement of the present invention, the step 1 specifically comprises:

[0013] Use flash / snapshot high-power VCSEL laser to emit pulsed laser to the target;

[0014] The pulsed laser is amplified to form emitted light and reaches the target location.

[0015] As a further improvement of the present invention, the step 2 specifically comprises:

[0016] receiving multiple beams of emitted light after being reflected by the target;

[0017] The laser receiver and the strobed imaging detector in the receiving system are adjusted as required to obtain a plurality of images with different depths of field.

[0018] As a further improvement of the present invention, the target is a moving target, and target detection is performed on the formed high-resolution three-dimensional image based on the LDA topic model to obtain the relative speed and relative distance of the moving target.

[0019] As a further improvement of the present invention, in the process of high-resolution three-dimensional imaging, large depth of field ranging is first used to achieve low-resolution target ranging; on the basis of the initial measurement, small depth of field three-dimensional imaging is used to obtain high-resolution three-dimensional data of the target, and at the same time in the continuous imaging process, the relative speed of the target is estimated according to the time differential of the distance, thereby achieving real-time high-precision three-dimensional imaging.

[0020] The present invention also discloses a multi-eye 3D laser radar imaging detection device, which is used to implement the multi-eye 3D laser radar imaging detection method, comprising:

[0021] A launching system for launching a laser at a target;

[0022] A receiving system is used to receive the emitted light after being reflected by the target to form images with different depths of field;

[0023] The data processing system is used to generate point cloud data based on images with different depths of field, and to extract features from the point cloud data based on a convolutional neural network to generate a high-resolution three-dimensional image; and to perform target detection on the high-resolution three-dimensional image based on an LDA topic model to generate a high-resolution three-dimensional image of the target.

[0024] As a further improvement of the present invention, the transmitting system comprises:

[0025] Pulsed laser power supplies for controlling flash / snapshot high-power VCSEL lasers;

[0026] Flash / snapshot high-power VCSEL laser, used to emit pulsed laser to the target under the control of pulsed laser power supply;

[0027] The transmitter is used to amplify the beam diameter of the pulsed laser to form emitted light and make the emitted light reach the target position.

[0028] As a further improvement of the present invention, the receiving system comprises:

[0029] A plurality of laser receivers, which are used to be adjusted according to requirements so as to receive a plurality of beams of emitted light reflected by a target;

[0030] A plurality of strobed imaging detectors corresponding to the plurality of laser receivers are used to form depth-of-field images for the corresponding emitted light to obtain a plurality of different depth-of-field images.

[0031] As a further improvement of the present invention, the data processing system comprises:

[0032] The image processing module is used to generate point cloud data according to the position of the laser receiver in the receiving system and the different depth of field images formed; the point cloud data is processed based on the convolutional neural network to complete feature extraction and form a high-resolution three-dimensional image;

[0033] The target detection module is used to perform target detection on high-resolution three-dimensional images based on the LDA topic model and generate high-resolution three-dimensional images of the target.

[0034] As a further improvement of the present invention, it also includes: a controller;

[0035] The controller is connected with the transmitting system, the receiving system and the data processing system, and is used to control the transmitting system, the receiving system and the data processing system to work in coordination.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The present invention adopts VCSEL and multi-CMOS three-dimensional imaging technology to solve the contradiction between high-resolution imaging, reliability and timeliness of flash / snapshot laser radar using indirect detection methods; at the same time, the entire device only adds multiple sets of optical imaging systems, which has better economic value than traditional solutions;

[0038] 2. The present invention adopts the large and small depth of field collaborative detection technology. First, a large depth of field is used for ranging to achieve low-resolution target ranging. On the basis of the initial measurement, a small depth of field three-dimensional imaging is used to obtain high-resolution three-dimensional data of the target. At the same time, in the continuous imaging process, the relative speed of the target is estimated according to the time differential of the distance, and the rough measurement link is reduced, thereby achieving real-time high-precision three-dimensional imaging. By processing radar images between different depths of field to form high-precision radar imaging, a large amount of computing resources are saved compared with traditional methods.

[0039] 3. The convolutional neural network adopted in the present invention has the characteristics of local receptive fields and shared weights. Compared with other neural networks, the parameters required for training of the convolutional neural network are greatly reduced; when performing high-speed target detection, the use of convolutional neural networks for target feature extraction can effectively reduce the delay and result deviation caused by data analysis; in the process of constructing high-precision three-dimensional images, the convolutional neural network algorithm is introduced to reduce the amount and complexity of data processing while retaining the effective and important features of the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of a multi-eye 3D laser radar imaging detection method disclosed in an embodiment of the present invention;

[0041] Figure 2 A framework diagram of a multi-eye 3D laser radar imaging detection device disclosed in an embodiment of the present invention;

[0042] Figure 3 An optical signal transmission route map of a multi-eye 3D laser radar imaging detection device disclosed in an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the physical detection principle of a multi-eye 3D laser radar imaging detection device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0046] like Figure 1 As shown, the present invention provides a multi-eye 3D laser radar imaging detection method, comprising:

[0047] S101, launching laser at the target;

[0048] Specifically include:

[0049] 1) Use flash / snapshot high-power VCSEL laser to emit pulsed laser or conventional laser to the target;

[0050] 2) The pulsed laser or conventional laser is amplified by the transmitter to form an emission light and reach the target position.

[0051] S102, controlling the receiving system to receive the emitted light reflected by the target to form images with different depths of field;

[0052] Specifically include:

[0053] 1) Adjust the receiver as required to receive multiple beams of emitted light reflected by the target;

[0054] 2) Adjust the laser receiver and the strobe imaging detector in the receiving system as required to obtain multiple images with different depths of field.

[0055] S103, generating point cloud data according to the images with different depths of field, and extracting features from the point cloud data based on a convolutional neural network to generate a high-resolution three-dimensional image;

[0056] Specifically include:

[0057] According to the position of the laser receiver in the receiving system and the different depth of field images formed, point cloud data is generated; the point cloud data is processed based on the convolutional neural network to complete feature extraction and form a high-resolution three-dimensional image; wherein,

[0058] Convolutional neural networks have characteristics such as local receptive fields and shared weights. Compared with other neural networks, the parameters that need to be trained for convolutional neural networks are greatly reduced. When performing high-speed target detection, using convolutional neural networks for target feature extraction can effectively reduce the delay and result deviation caused by data analysis.

[0059] S104, performing target detection on the high-resolution three-dimensional image based on the LDA topic model, obtaining the relative speed and relative distance of the moving target, and generating a high-resolution three-dimensional image of the target;

[0060] in,

[0061] The category of the target to be detected is taken as the theme, and a model of target category and features is constructed. The above categories include two different categories, target and non-target, as well as different directions and different spectral intensities. The above features are mapped from the Bag-of-words model.

[0062] The Bag-of-words model is described as follows:

[0063] ① For an image, a Dirichlet distribution with α as a hyperparameter is used to generate the distribution θ of the target category; ② For each region, a multinomial distribution with θ as a parameter is used to sample and generate a target category z i ; ③ Given a target category, sample word w through Dirichlet distribution with β as hyperparameter i , where words are mapped from image features by the Bag-of-Words model.

[0064] In the process of high-resolution three-dimensional imaging, large depth of field ranging is first used to achieve low-resolution target ranging; based on the initial measurement, small depth of field three-dimensional imaging is used to obtain high-resolution three-dimensional data of the target. At the same time, in the continuous imaging process, the relative speed of the target is estimated based on the time differential of the distance, thereby reducing the rough measurement links and achieving real-time high-precision three-dimensional imaging.

[0065] Traditional flash / snapshot laser radars that use indirect detection methods cannot effectively balance the contradiction between high-precision, timeliness and reliability of images in high-precision radar imaging. The use of dual radars to construct high-precision images brings great economic costs, and the traditional image processing and recognition process either cannot guarantee the extraction of effective and important features, or requires the processing of a large amount of redundant information. To this end, this patent uses multi-CMOS three-dimensional imaging technology, and multiple eyes perform data detection simultaneously, which reduces data collection time, improves the timeliness of the receiving stage, and provides a basis for clear imaging of high-speed targets.

[0066] like Figure 2-4 As shown, the present invention provides a multi-eye 3D laser radar imaging detection device, which is used to implement the above-mentioned multi-eye 3D laser radar imaging detection method, including: a transmitting system 1, a receiving system 2, a data processing system 3 and a controller 4; wherein,

[0067] The emission system 1 of the present invention is used to emit laser to a target 5; wherein the emission system 1 comprises: a pulse laser power supply 13, a flash / snapshot type high-power VCSEL laser 11 and a transmitter 12, the pulse laser power supply 13 controls the flash / snapshot type high-power VCSEL laser 11 to emit pulse laser to the target, the transmitter 12 amplifies the beam diameter of the pulse laser to form emission light, and makes the emission light reach the target position.

[0068] The receiving system 2 of the present invention is used to receive the emitted light after being reflected by the target to form images with different depths of field; wherein the receiving system 2 includes a plurality of laser receivers 21 and a plurality of gated imaging detectors 22 corresponding to the plurality of laser receivers 21, and the laser receivers 21 are adjusted as required to receive the plurality of emitted light beams after being reflected by the target; the plurality of gated imaging detectors 22 form depth of field images for the corresponding emitted light to obtain a plurality of images with different depths of field. Furthermore, the gated imaging detector can also be a CMOS image sensor.

[0069] The data processing system 3 of the present invention includes an image processing module 31 and a target detection module 32. The image processing module 31 is used to generate point cloud data according to the position of the laser receiver in the receiving system and the different depth of field images formed; the point cloud data is processed based on a convolutional neural network to complete feature extraction and form a high-resolution three-dimensional image; the target detection module 32 is used to perform target detection on the high-resolution three-dimensional image based on the LDA topic model to generate a high-resolution three-dimensional image of the target.

[0070] The controller 4 of the present invention is connected to the transmitting system 1, the receiving system 2 and the data processing system 3, and is used to control the transmitting system 1, the receiving system 2 and the data processing system 3 to work in coordination.

[0071] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-eye 3D laser radar imaging detection method, characterized in that: include: Step 1: Use a flash / snapshot high-power VCSEL laser to emit pulsed laser to the moving target. The pulsed laser is amplified to form emission light and reaches the target position. Step 2, receiving the emitted light reflected by the target to form images with different depths of field; Step 3: Generate point cloud data according to the position of the laser receiver in the receiving system and the different depth of field images formed, and perform feature extraction on the point cloud data based on the convolutional neural network to generate a high-resolution three-dimensional image; wherein the receiving system includes a plurality of laser receivers and a plurality of gated imaging detectors corresponding to the plurality of laser receivers, and the laser receivers are adjusted according to requirements to receive a plurality of beams of emitted light reflected by the target; the plurality of gated imaging detectors form depth of field images for the corresponding emitted light to obtain a plurality of different depth of field images; Step 4: Detect the target on the formed high-resolution three-dimensional image based on the LDA topic model, obtain the relative speed and relative distance of the moving target, and generate a high-resolution three-dimensional image of the target; specifically, the following steps are included: The category of the target to be detected is used as the theme to construct a model of target category and features; the above categories include both target and non-target categories, as well as different directions and different spectral intensities. The above features are mapped from the Bag-of-words model; The Bag-of-words model is described as follows: ① For an image, a Dirichlet distribution with α as a hyperparameter is used to generate the distribution θ of the target category; ② For each region, a multinomial distribution with θ as a parameter is sampled to generate a target category z i ; ③ Given a target category, sample word w through Dirichlet distribution with β as hyperparameter i , where words are mapped from image features by the Bag-of-Words model; In the process of high-resolution three-dimensional imaging, large depth of field ranging is first used to achieve low-resolution target ranging; based on the initial measurement, small depth of field three-dimensional imaging is used to obtain high-resolution three-dimensional data of the target. At the same time, in the continuous imaging process, the relative speed of the target is estimated according to the time differential of the distance, thereby achieving real-time and high-precision three-dimensional imaging.

2. The multi-eye 3D laser radar imaging detection method according to claim 1, characterized in that: The step 2 specifically includes: receiving multiple beams of emitted light after being reflected by the target; The laser receiver and the strobed imaging detector in the receiving system are adjusted as required to obtain a plurality of images with different depths of field.

3. A multi-eye 3D laser radar imaging detection device, used to implement the multi-eye 3D laser radar imaging detection method as described in any one of claims 1 to 2, characterized in that: include: The launch system is used to launch pulsed lasers at moving targets using flash / snapshot high-power VCSEL lasers. The pulsed lasers are amplified to form emission light and reach the target location. A receiving system is used to receive the emitted light after being reflected by the target to form images with different depths of field; wherein the receiving system includes a plurality of laser receivers and a plurality of strobed imaging detectors corresponding to the plurality of laser receivers, and the laser receivers are adjusted according to requirements to receive the plurality of emitted light beams reflected by the target; the plurality of strobed imaging detectors form depth of field images for the corresponding emitted light to obtain a plurality of images with different depths of field; The data processing system is used to generate point cloud data according to the position of the laser receiver in the receiving system and the different depth of field images formed, and to extract features of the point cloud data based on the convolutional neural network to generate a high-resolution three-dimensional image; to perform target detection on the formed high-resolution three-dimensional image based on the LDA topic model, to obtain the relative speed and relative distance of the moving target, and to generate a high-resolution three-dimensional image of the target; specifically includes: The category of the target to be detected is used as the theme to construct a model of target category and features; the above categories include both target and non-target categories, as well as different directions and different spectral intensities. The above features are mapped from the Bag-of-words model; The Bag-of-words model is described as follows: ① For an image, a Dirichlet distribution with α as a hyperparameter is used to generate the distribution θ of the target category; ② For each region, a multinomial distribution with θ as a parameter is sampled to generate a target category z i ; ③ Given a target category, sample word w through Dirichlet distribution with β as hyperparameter i , where words are mapped from image features by the Bag-of-Words model; In the process of high-resolution three-dimensional imaging, large depth of field ranging is first used to achieve low-resolution target ranging; based on the initial measurement, small depth of field three-dimensional imaging is used to obtain high-resolution three-dimensional data of the target. At the same time, in the continuous imaging process, the relative speed of the target is estimated according to the time differential of the distance, thereby achieving real-time and high-precision three-dimensional imaging.

4. The multi-eye 3D laser radar imaging detection device according to claim 3, characterized in that: The transmitting system comprises: Pulsed laser power supplies for controlling flash / snapshot high-power VCSEL lasers; Flash / snapshot high-power VCSEL laser, used to emit pulsed laser to the target under the control of pulsed laser power supply; The transmitter is used to amplify the beam diameter of the pulsed laser to form emitted light and make the emitted light reach the target position.

5. The multi-eye 3D laser radar imaging detection device according to claim 3, characterized in that: Also includes: Controller; The controller is connected with the transmitting system, the receiving system and the data processing system, and is used to control the transmitting system, the receiving system and the data processing system to work in coordination.

Images