A noise suppression method based on multiplexing in non-visual target detection

Abstract

The application provides a noise suppression method based on multiplexing in non-visual field target detection. The method divides the return signal into multiple signals by means of a beam splitter, makes the derived multiple signals have time correlation characteristics, uses the characteristics to evaluate the confidence of the echo signal, retains and amplifies the echo signal with time correlation characteristics, suppresses random noise in the time domain, suppresses post-pulse noise caused by high-count primary echo, reduces the false alarm probability of detection, and thus effectively extracts the target signal. The method can be applied to a collection system, does not need additional data processing equipment to align the starting time of the multiple detectors, can suppress post-pulse noise, expand the detection range of the target, filter out random noise, reduce the false alarm probability, and is beneficial to subsequent estimation of the position of the hidden target.

Description

Technical Field

[0001] This invention belongs to the field of photoelectric detection, specifically relating to a noise suppression method based on multiplexing in non-line-of-sight target detection. Background Technology

[0002] Non-line-of-sight (NLOS) detection technology is a relatively new technology proposed in recent years. It can detect, track, or image scenes obscured by obstacles, extending the range of human vision. In the future, it has significant potential applications in areas such as safe autonomous driving, emergency rescue, and robot vision. It has become a hot research topic both domestically and internationally. Unlike traditional optical detection technology, which can only detect objects within the line of sight and cannot detect objects if the target is obstructed, NLOS technology offers a different approach.

[0003] This technology employs active detection, emitting a pulsed laser beam onto an intermediate surface. The laser undergoes diffuse reflection at the intermediate surface; a portion of the laser returns directly to the detector and is collected, recorded as the first echo. The remaining diffusely reflected laser strikes the hidden target and undergoes a second diffuse reflection back to the intermediate surface, where a third diffuse reflection occurs. The photons returning after this third reflection and being collected are called the third echo. The distance traveled by the photons is measured, allowing the calculation of information about the hidden target and enabling the detection of targets outside the field of view. However, due to the multiple diffuse reflections during detection, the histogram signal of the photons returning from the target is extremely weak, often drowned out by noise, making the target signal difficult to detect.

[0004] From the perspective of noise types in the photon arrival time histogram, it generally includes ambient noise outside the detector, dark count noise inside the detector, and afterpulse noise. Ambient noise can generally be suppressed to some extent by background subtraction methods, and this background signal can also be estimated by averaging the signals from the previous few frames. However, some noise that is difficult to remove still exists and affects the identification of the target signal. Dark count noise inside the detector is closely related to temperature. With current technology, dark count noise can be less than 1kHz, which is far less than the impact of ambient noise. Afterpulse noise is caused by the primary echo returning directly from the intermediate surface containing a large number of photons. The detector will undergo secondary detection, resulting in increased noise, decreased detection probability, and increased false alarm probability. By using a single-photon detector with gated mode, the high-intensity primary echo can be shielded, thereby avoiding the impact of afterpulse noise caused by the primary echo. However, for single-photon detectors operating in free mode, afterpulse noise caused by the primary echo is unavoidable. In general, using a single-channel single-photon detector to detect echo photons is easily affected by random noise and afterpulse noise, making it difficult to detect and identify the target signal. Summary of the Invention

[0005] To address the problem that the echo signal returned by hidden targets in non-line-of-sight target detection is extremely weak and easily affected by noise, making the target signal difficult to detect and identify, this invention provides a noise suppression method based on multiplexing for non-line-of-sight target detection. The technical solution adopted by the method is as follows:

[0006] First, the echo signal is split into multiple signals using a beam splitter, giving the derived signals time-dependent characteristics. Then, this characteristic is used to assess the confidence level of the echo signal, retaining and amplifying the echo signal with time-dependent characteristics, and suppressing random noise and after-pulse noise in the time domain, thereby effectively extracting the target signal.

[0007] Furthermore, the method of splitting the echo signal into multiple paths by means of a beam splitter includes splitting the collected return photon beam into multiple beams. The beam splitting method includes using a beam splitter mirror or an optical fiber splitter, the purpose of which is to split the beam into multiple beams according to energy.

[0008] Furthermore, the confidence level of the echo signal is assessed using this feature, including the following steps:

[0009] Step 1: Time-shift the photon arrival time histogram of the multi-channel array based on the highest peak to align the delays caused by different detector response times;

[0010] Step 2: Preprocess the photon arrival time histogram, including zeroing the first echo in the photon arrival time histogram, subtracting background noise, filtering and smoothing operations;

[0011] Step 3: Perform a minimum hold operation on each time resolution unit in the input multi-channel photon arrival time histogram to obtain the confidence coefficients of the echoes from multiple channels.

[0012] Step 4: Calculate the Hadamard product of the confidence coefficient and the photon arrival time histograms of multiple channels.

[0013] The advantages of this invention compared to the prior art are as follows:

[0014] (1) No additional data processing equipment is required to align the start times of the multi-detectors;

[0015] (2) It can suppress after-pulse noise and expand the detection area of ​​the target;

[0016] (3) Filtering out random noise reduces the probability of false alarms and is beneficial for subsequent estimation of the location of hidden targets. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the data acquisition device used in the non-line-of-sight target detection experiment of this invention;

[0018] Figure 2 These are the original photon arrival time histograms for channel one and channel two in this invention;

[0019] Figure 3 This is the photon arrival time histogram after preprocessing in channel one of this invention;

[0020] Figure 4 This is the photon arrival time histogram after preprocessing in channel two of this invention;

[0021] Figure 5 This is a confidence coefficient diagram of the channel one photon arrival time histogram and the channel two photon arrival time histogram after minimum preservation in this invention;

[0022] Figure 6 This is the result diagram after noise suppression in this invention. Detailed Implementation

[0023] The following are specific embodiments of the present invention. However, the following examples are only for explaining the present invention, and the scope of protection of the present invention should include all the contents of the claims. Moreover, those skilled in the art can implement all the contents of the claims of the present invention through the following examples.

[0024] This embodiment uses a photon arrival time histogram acquired using a two-channel structure in a non-line-of-sight target detection experiment as an example. A schematic diagram of the acquisition device is shown below. Figure 1 As shown. The specific steps of this invention are as follows:

[0025] First, a pulsed laser is emitted towards the intermediate surface. The first echo directly returning from the intermediate surface and the third echo returning from the hidden target are collected. Echo signals returning from different positions on the intermediate surface are collected using a galvanometer and split into two beams by a beam splitter. These beams are then independently detected using two channels of a single-photon detector system. In this embodiment, a beam splitter is used to split the returning beam into two beams, but the method is not limited to two beams. The photon arrival time histograms for the two channels are shown below. Figure 2 As shown, the total length of each collected frame of signal is 25 ns. Then, the signal is discretized into 2500 time-resolved units in the time dimension at intervals of 0.01 ns.

[0026] Then, the minimum preservation of multiple signals and the Hadamard product of multiple signals are calculated to preserve and amplify the echo signal with time-dependent characteristics, suppress random noise and after-pulse noise, thereby extracting the target signal. This includes the following steps:

[0027] Step 1: Time-shift the photon arrival time histogram of the multi-channel array based on the highest peak to align the delays caused by different detector response times. Figure 2The highest peak of the signal shown represents the position of the first echo that the pulsed laser returns directly from the intermediate surface. Because its count value is high and easy to identify, the highest peak is used as the reference time for aligning the photon arrival time histogram to align the delay caused by the different detector response times.

[0028] Step 2: Preprocess the photon arrival time histogram, including zeroing the first echo in the photon arrival time histogram, subtracting background noise, filtering and smoothing operations. Figure 2 In the photon arrival time histogram of the two channels, a high first echo count can obscure the third echo signal returned by the hidden target. Therefore, we set the first echo portion of the signal to zero to remove it. We also subtracted the acquired background signal to reduce background noise. The processed photon arrival time histogram of the two channels is shown below. Figure 3 , Figure 4 As shown. But it is not limited to using any method to subtract background noise. Then, we perform one-dimensional mean filtering on the signal in the time dimension and obtain a smoother signal. The mean filtering is shown in equation (1). Here, it is not limited to using the mean filtering method.

[0029]

[0030] Where h(i) represents the filtered value at each time resolution unit, f(i) represents the initial value at the i-th time resolution unit, k is the subscript of the time resolution unit within the filtering window, L is the length of the filtering window, and N is the total length of the histogram.

[0031] Step 3: Perform a minimum preserve operation on each time-resolution unit in the input multi-channel photon arrival time histogram to obtain the confidence coefficients of the echoes from multiple channels. Figure 3 , Figure 4 In the histogram, we can see that even after subtracting the background from the histogram, we cannot determine which echoes originate from the hidden object. Furthermore, we note that... Figure 4 There is strong noise at approximately 19 ns, which affects the identification of the three echoes and is difficult to remove. Figure 5 In this process, we perform minimum preservation of the histograms of the two channels, as shown in equation (2).

[0032]

[0033] Where X1 and X2 are the photon arrival time histograms of the two channels, W is the output with minimum hold, and i represents each time resolution unit of the histogram. Figure 5As shown, this operation performs a minimum hold on the input to calculate the minimum count of the histograms of the two channels, thereby obtaining the confidence coefficients of the echoes of the two channels, since the signal portion has time-dependent characteristics, while the noise is randomly generated. Then, the portions with confidence values ​​less than 0 are set to zero, indicating that the target signal does not exist there.

[0034] Step 4: Calculate the confidence coefficient and the Hadamard product of the photon arrival time histograms of multiple channels. The confidence coefficient and the Hadamard product of the two channels can be understood as a weighted calculation of the input signals. The Hadamard product of the two input channels amplifies the similar parts of the two inputs, while minimum preservation provides the confidence coefficient, as shown in equation (3):

[0035] Y=W⊙X1⊙X2 (3)

[0036] Where X1 and X2 are the acquired photon arrival time histograms from the two channels, W is the minimum hold output, i represents each time resolution unit of the histogram, and ⊙ is the Hadamard product. Thus, the processed photon arrival time histogram signal is obtained, as follows: Figure 6 As shown. After calculation and processing, it can be obtained from... Figure 6 Clearly identifying the location of the target signal is beneficial for subsequent non-line-of-sight detection, tracking, or reconstruction.

[0037] In this embodiment, the returned photon beam during non-line-of-sight detection is split into two channels of the single-photon detection system for independent detection. The confidence level of the photon arrival time histograms output by multiple detectors is evaluated and the signal is amplified. By utilizing the time correlation characteristics of the multi-channel system, random noise and afterpulse effects are suppressed, and the target signal is extracted, ensuring the smooth progress of subsequent processing.

[0038] The parts of this invention not described in detail are well-known to those skilled in the art.

Claims

1. A noise suppression method based on multiplexing in non-line-of-sight target detection, characterized in that, The echo signal is split into multiple signals using a beam splitter, giving the derived signals time-correlation characteristics. These time-correlation characteristics are then used to assess the confidence level of the echo signals. The echo signals with time-correlation characteristics are retained and amplified, while random noise and after-pulse noise in the time domain are suppressed, thereby effectively extracting the target signal. The confidence assessment of the echo signals using the time-correlation characteristics includes the following steps: Step 1: Time-shift the photon arrival time histogram of the multi-channel array based on the highest peak to align the delays caused by different detector response times; Step 2: Preprocess the photon arrival time histogram, including zeroing the first echo in the photon arrival time histogram, subtracting background noise, filtering and smoothing operations; Step 3: Perform a minimum hold operation on each time resolution unit in the input multi-channel photon arrival time histogram to obtain the confidence coefficients of the echoes from multiple channels; Step 4: Calculate the Hadamard product of the confidence coefficient and the photon arrival time histograms of multiple channels to obtain the processed photon arrival time histogram signal.

2. The noise suppression method based on multiplexing in non-line-of-sight target detection according to claim 1, characterized in that: The method of splitting the echo signal into multiple signals by means of a beam splitter includes splitting the collected return photon beam into multiple beams, and the beam splitting method includes using a beam splitter mirror or an optical fiber splitter.

Images