Silicon Micro implements low-computing, low-memory millimeter-wave radar target detection solution

At present, the field of target detection mainly relies on infrared sensors, 5.8G microwave radar and other sensor technologies to collect data, and then use relevant algorithms to detect and extract corresponding targets.

In these implementation schemes, infrared sensors are greatly affected by ambient temperature. When the temperature is high or the application environment temperature exceeds room temperature (for example, when integrated inside a lighting bulb or inside a chassis), the detection sensitivity and accuracy decrease. The 5.8G microwave radar technology mainly uses the Doppler frequency shift principle to detect moving targets, but cannot detect relatively stationary micro-movement targets.

In response to the problems with the above-mentioned sensors, existing technologies use methods such as combining cameras with machine learning to identify the human body, but the cost is high and the privacy and reliability are low. In particular, installing cameras in home spaces can easily violate privacy, and there are also fewer applicable scenarios.

Therefore, on March 24, 2020, Silicon Micro applied for an invention patent entitled "Target detection method and device, electronic device, and storage medium based on millimeter-wave radar" (application number: 202010213379.X), and the applicant was Nanjing Silicon Microsystems Co., Ltd.

Based on the relevant information currently disclosed by the patent, let us take a look at this technical solution.

As shown in the figure above, it is a flow chart of the target detection method based on millimeter-wave radar invented in this patent. First, the system will pre-process the millimeter-wave radar echo data received at the current moment. Since the radar echo data is a single-channel data that is sent and received, this data contains a packet header and the data itself, so it is necessary to pre-process it first. The pre-processing operation includes verifying the radar echo data received at the current moment, extracting a single frame of data therein and performing a fast Fourier transform.

Secondly, the preprocessed radar echo data is background-cancelled with the historical radar echo data to obtain fast-time data. That is, by accumulating and averaging the radar echo data within a period of time before the current moment, the signal-to-noise ratio can be effectively improved. After filtering for this period of time, relatively clean background echo data can be obtained.

Next, MTI cancellation is performed on the fast time data to determine whether there is a dynamic target. If not, the mean square error of the phase signal in the range threshold signal corresponding to the dynamic target within the first preset time length after the last fast time appearance is calculated. Specifically, the range threshold signal corresponds to the fast time data, and each discrete sampling point in it represents a range gate.

When a target enters the detection area, the signal level quickly rises to a high level state and remains at the high level state when the target is in the detection area; after the target appears in the detection area for the last time, the corresponding distance threshold signal is continuously sampled for the first preset time length, so as to determine whether the target is in a micro-motion state or has left the detection area, as shown in the following figure:

Taking a bandwidth of 1 GHz as an example, the corresponding distance resolution is 15 cm, that is, each sampling point of the discrete signal in fast time represents a distance of 15 cm. If a target feature is found at the 10th sampling point, it means that there is a target at a distance of 150 cm. By sampling the sampling point for the first preset time, it can be determined whether there is a micro-moving target (T2-T3) based on the fluctuation of the phase signal of the distance threshold signal during this time.

Finally, the mean square error is calculated to determine whether there is a micro-moving target at the current moment. A calculation threshold is set in the system, and its value is determined according to the actual hardware and transmit and receive power of the application. If the mean square error calculated by judgment is greater than the preset threshold, it is determined that there is a micro-moving target at the current moment.

The above is the target detection solution based on millimeter-wave radar invented by Silicon Micro. While ensuring the detection accuracy, this solution uses an algorithm with low computing power requirements, so that the solution can be applied to low-cost, low-memory and low-frequency processors. There is no need to configure hardware facilities such as cameras, and the privacy is also good.