A non-visual target detection method based on blind source separation

Abstract

The application provides a non-visual field target detection method based on blind source separation, and solves the problem of target signal aliasing in non-visual field detection, that is, the echo signals between the target and the background and multiple targets are difficult to distinguish in some scenes. The application first transmits laser of multiple spectral segments to an intermediate surface, then selects and receives echo of different spectral segments by a detector, and then processes the echo signals of each spectral segment by a blind source separation method, so as to realize separation and detection of each non-visual field target. The application has the beneficial effects that: under the condition of lacking prior information of the target and the background, the non-visual field multiple target separation or the extraction of the weak target from the complex background can be realized; neither the background needs to be collected in advance, nor the target needs to move, and the distance and shape between the targets are not limited; the application is suitable for both non-visual field tracking and non-visual field imaging.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of photoelectric detection, and particularly relates to a non-visual field target detection method based on blind source separation. BACKGROUND

[0002] Non-visual field (NLOS) detection technology is a new technology for optical detection of hidden targets outside the line of sight, similar to "bending the line of sight" or "watching through a wall". Traditional detection technology can usually only detect targets within the line of sight (visual field), and the information of targets outside the line of sight (non-visual field) has important value in some special environments. If the movement of criminals can be observed by using NLOS detection technology, targeted deployment can be made to reduce the danger of the task. For example, in automatic driving, NLOS detection can expand the line of sight range and avoid accidents. For example, in fire rescue, NLOS detection can predict the position and situation of personnel in advance, facilitating rescue. Therefore, NLOS detection technology has broad application prospects in the fields of robot vision, manufacturing, medical imaging, automatic driving, and exploration and rescue in extreme conditions.

[0003] The current common method of NLOS detection is to emit pulsed laser to the diffuse reflection interface (such as wall, ground, door panel) near the target, and measure the echo scattered to the target to obtain the photon flight time and flight distance, and then obtain the spatial information of the target through multi-point scanning or detector array, so as to realize the tracking or imaging of the target, as shown in Figure 1

[0004] The research of NLOS detection technology mainly has two directions of tracking and imaging. NLOS tracking only focuses on the position of the hidden target; NLOS imaging is committed to reconstructing the three-dimensional surface shape of the hidden target. The current research mainly improves the detection device, scanning method and reconstruction algorithm, and a series of effective detection methods emerge.

[0005] However, whether it is NLOS tracking or imaging, there is a common problem at present - target signal aliasing, that is, the echo signals between the target and the background and multiple targets are difficult to distinguish in some scenes.

[0006] Firstly, because the target signal is very weak, it is easy to be confused with the background, and it is difficult to distinguish the target and the background without certain prior knowledge. As shown in Figure 2 The largest wave peak is formed by the photons directly returned from the interface. In order to observe the weak target signal, Figure 2 ​(b) in the figure shows the amplified original signal, the target position signal is marked with "O", and the intermediate surface echo is marked with "X". It can be seen that the target signal is very weak, and even without the intermediate surface echo, it is almost submerged in the background. Figure 2 (c) in the figure is the extracted target signal after background subtraction processing. It can be seen that the target has only a few photons. In existing algorithms, in order to implement effective background subtraction processing, it is often necessary to pre-acquire the waveform of the background or estimate the background waveform in the case of target motion. However, in practical applications, these methods have great limitations. It is impossible to predict the use site in advance, and it is also impossible to guarantee that the target is moving. Therefore, the background subtraction extraction method faces the bottleneck of practicality.

[0007] In addition, if there are multiple targets in the scene, targets that are close to each other are also easily confused together. As shown in Figure 3 When two targets are close to each other, their signals will overlap, making it difficult to distinguish them, and further causing target extraction errors or loss.

[0008] The problem of target signal overlap poses a great obstacle to the practicality of NLOS detection technology. Since the scene in reality is relatively complex, such as detecting pedestrians behind a corner, various facility interference and multiple people in parallel are inevitable. Therefore, if the correct target cannot be identified from the background, the NLOS detection loses its meaning.

[0009] For the problem of target signal overlap in NLOS detection, there is no good solution in existing literature. Most of the literature only avoids this problem by constructing a single scene; some literature uses background subtraction to separate the target from the background, but requires pre-acquisition of the background or movement of the target; some literature distinguishes different targets by their position, shape, reflectivity and other characteristics, but the targets must be kept at a certain distance, and the application is limited. Therefore, the problem of target signal overlap remains unresolved and urgently needs a new breakthrough. SUMMARY

[0010] The technical problem solved by the present application is the problem of target signal overlap in non-line-of-sight target detection, that is, the problem of distinguishing the echo signals between the target and the background and multiple targets in some scenes.

[0011] Summary of the application: the application provides a non-visual target detection method based on blind source separation, which utilizes the principle that targets of different materials have different spectral reflectivity, first emits laser of multiple spectral bands to an interface, then uses a detector to select and receive echoes of different spectral bands, acquires echo photon flight time statistical signals (referred to as "echo signals") of each spectral band, processes the echo signals of each spectral band by using a blind source separation method, obtains signal families belonging to different targets, finally implements a non-visual detection algorithm on the signal families of each target, and further realizes separation and detection of each non-visual target. The technical scheme provided by the application is as follows:

[0012] A non-visual target detection method based on blind source separation, the method comprising the following steps:

[0013] Step (1): emitting laser of multiple spectral bands to an interface;

[0014] Step (2): using a detector to select and receive echoes of different spectral bands;

[0015] Step (3): processing echo signals of each spectral band by using a blind source separation method, and further realizing separation and detection of each non-visual target.

[0016] Further, in step (1), when emitting laser of multiple spectral bands to an interface, all the laser is emitted to the same point on the interface.

[0017] Further, in step (2), when using a detector to select and receive echoes of different spectral bands, a plurality of narrow-band optical filters corresponding to the spectral bands are used to realize selection of the spectral bands, so that each detector only receives echoes of a single spectral band scattered by a target.

[0018] Further, in step (3), when processing echo signals of each spectral band by using a blind source separation algorithm, an independent component analysis algorithm is used.

[0019] The application has the following beneficial effects compared with the prior art:

[0020] (1) The non-visual multi-target separation or extraction of weak targets from a complex background can be realized under the condition of lacking prior information of targets and backgrounds;

[0021] (2) Neither pre-acquisition of backgrounds nor target movement is needed, and the method is not limited by distances and shapes of targets;

[0022] (3) The method is suitable for NLOS tracking and NLOS imaging. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a schematic diagram of the basic principle of NLOS detection;

[0024] Figure 2 is an example diagram of echo signals of NLOS;

[0025] Figure 3 is an example diagram of signal aliasing when two targets are close;

[0026] Figure 4 is a schematic diagram of an embodiment of the embodiment;

[0027] Figure 5 is a schematic diagram of a blind source separation processing framework of the embodiment. DETAILED DESCRIPTION

[0028] The following is a specific implementation of the present application. However, the following embodiments are only for the purpose of explaining the present application, and the scope of protection of the present application should include the entire content of the claims, and the skilled in the art can achieve the entire content of the claims of the present application through the following embodiments.

[0029] The non-line-of-sight target detection method based on blind source separation provided by the present application mainly includes the following steps:

[0030] Step (1): Emit laser of multiple spectral bands to the intermediate surface;

[0031] Step (2): Use the detector to select the echo of different spectral bands;

[0032] Step (3): Use the blind source separation method to process the echo signals of each spectral band, and then realize the separation and detection of each non-line-of-sight target.

[0033] This embodiment takes the detection of a maximum of 3 targets by 3 spectral bands as an example for illustration, and the schematic diagram of the implementation scheme is as shown in Figure 4 .

[0034] In step (1), first, in terms of light source, since the present application introduces spectral information to expand the signal dimension of target echo, a multi-spectral pulsed laser is used as the light source. In order to meet the conditions of blind source separation, the number of spectral bands is generally not less than the number of targets. Therefore, this embodiment uses 3 spectral bands to detect a maximum of 3 targets.

[0035] In terms of spectral band selection of the laser, this embodiment uses three spectral bands of 1550nm, 1064nm and 850nm, which is conducive to avoiding the occurrence of a sick matrix, and also conducive to filtering out ambient light.

[0036] In terms of laser emission mode, in order to avoid the time-of-flight error of photons of different spectral bands, all lasers need to be emitted to the same point on the intermediate surface, thereby ensuring that the subsequent scattering paths are consistent. Therefore, calibration needs to be performed on several lasers before implementation, so that they are aligned to the same origin on the intermediate surface.

[0037] In other parameters of the laser, the common parameters of the existing NLOS detection are used, such as pulse frequency 10-40 MHz, average power 100-1000 mW, pulse width 50-100 ps, etc. But in order to calculate the photon flight time, the lasers need to work synchronously. Therefore, the output synchronization signal of one laser is used to make the other two lasers work in the external synchronization mode.

[0038] In step (2), in terms of echo receiving devices, the SNSPD is selected as the detector in the embodiment. The detection efficiency of SNSPD in the spectral segment 800-1600 nm is greater than 70%, which is very suitable for multi-spectral receiving devices. A set of SNSPD can integrate multiple detectors, which can simultaneously detect multiple groups of signals. The number of detectors used in practice is determined according to the requirements of the NLOS detection algorithm, such as: the back projection tracking algorithm can use more than 3 detectors to form an array to fuse positioning, and the confocal imaging algorithm only needs 1 detector, but needs to use multiple points to scan the interface.

[0039] In terms of receiving optical systems, a long-focus optical lens is placed in front of each detector in the embodiment, which is used to collect echo photons. If the distance is far, a telescope can be used instead of an optical lens. In addition, the important difference between the present application and the existing method is that multiple narrow-band filters corresponding to the spectral segments are used to realize the selection of the spectral segments. A filter disc is placed between each detector and the optical lens, which is used to dynamically select different filters to receive photons of different wave bands. The filter corresponds one-to-one to the spectrum of the laser, so that the detector only receives the echo of the target scattered single spectral segment.

[0040] In terms of photon flight time acquisition, a single photon counter is used to count the relationship curve between the number of photons and the flight time. The single photon counter takes the electric pulse synchronously emitted by the laser pulse as the time zero point, and records the time distribution of the number of photons measured by the single photon detector. The photon flight time statistical graph output by the single photon counter is the input signal of the NLOS detection algorithm.

[0041] In step (3), in terms of signal processing, the present application proposes as Figure 5The shown blind source separation processing framework, wherein the blind source separation algorithm is implemented by independent component analysis (ICA) algorithm. First, for each detector or scanning point, the photon time-of-flight distribution signals of 3 different spectral segments are received by adjusting the filter disc, thereby expanding the signal dimension of the target echo. Since the signals of different spectral segments are the mixed echoes of each target source signal with different reflectivity, and the different reflectivity of different spectral segments makes the mixing coefficients different, it conforms to the mathematical model of ICA. Send it into the ICA algorithm for separation, and three independent signals can be obtained. Each detector or scanning point will obtain three independent signals. Then, the signal groups belonging to different targets (from all detectors or scanning points) are sent into the NLOS detection algorithm respectively, and the detection results of each target are obtained.

[0042] In terms of ICA algorithm, the embodiment adopts FastICA algorithm. The advantage of FastICA is that the algorithm is mature, the calculation speed is fast, and each target can be extracted one by one, that is, only the target of interest can be extracted, without separating all signals. Since the FastICA algorithm takes the non-Gaussianity of the signal as the independent evaluation criterion, and the target often has stronger non-Gaussianity than the background, the algorithm is expected to extract the target signal preferentially in a complex background, thereby solving the problem of separating the target under unknown background conditions. The specific implementation steps of the FastICA algorithm are not described here.

[0043] In terms of NLOS detection algorithm, the embodiment adopts the NLOS tracking algorithm based on neural network or the NLOS imaging algorithm based on LCT to track or image each target. The specific implementation steps of these algorithms are not described here.

[0044] The part of the present application not described in detail belongs to the known technology of those skilled in the art.

Claims

1. A non-line-of-sight target detection method based on blind source separation, characterized in that, The method includes the following steps: Step (1): Emit lasers of multiple spectral bands to the intermediate surface; Step (2): Use a detector to select and receive echoes from different spectral bands; Step (3): Use the blind source separation method to process the echo signals of each spectral band, thereby realizing the separation and detection of each non-line-of-sight target; In step (1), when emitting lasers of multiple spectral bands to the intermediate surface, all lasers are emitted to the same point on the intermediate surface. In step (2), when the detector selects to receive echoes of different spectral bands, multiple narrowband filters corresponding to the spectral bands are used to achieve the selection of the spectral bands, so that each detector only receives the echoes of a single spectral band scattered by the target. In step (3), when processing the echo signals of each spectral band using the blind source separation method, the independent component analysis algorithm is used. First, for each detector or scanning point, the photon time-of-flight distribution signals of three different spectral bands are received by adjusting the filter disc, thereby expanding the signal dimension of the target echo. Since the signals of different spectral bands are the mixture of the echoes scattered by the signals of each target source with different reflectivities, and the different reflectivities of different spectral bands result in different mixing coefficients, this conforms to the mathematical model of the Independent Component Analysis (ICA) algorithm. The ICA algorithm is then fed into the ICA algorithm for separation, resulting in three independent signals. Each detector or scanning point will receive three independent signals. Then, the signal families belonging to different targets are fed into the NLOS detection algorithm to obtain the detection results of each target.

Images