Intelligent mobile garbage can with voice control and remote control functions and control method thereof

Abstract

The invention relates to the technical field of garbage cans, and discloses an intelligent mobile garbage can with voice control and remote control functions and a control method thereof.The intelligent mobile garbage can comprises a garbage can body, a can cover is installed on the top of the garbage can body, and an annular microphone array is installed on the front side of the garbage can body; the garbage can comprises a garbage can body, a TOF camera is installed on the front side of the garbage can body, an ultrasonic sensor is installed outside the garbage can body, a buzzer alarm is installed on the left side of the garbage can body, and a magnetic type charging contact is installed at the bottom of the garbage can body. According to the invention, the sound source orientation precision of + / -5 degrees is realized through the annular microphone array + TDOA algorithm, the instruction recognition rate of 92% or above is still kept in a noise environment in combination with the DPRNN model, the false triggering probability is remarkably reduced, dual identity verification: voiceprint comparison and off-line / on-line instruction analysis dual verification is carried out, illegal user control or pet false triggering is prevented, and the safety of a pet is improved. And the safety is improved by more than 80%.

Description

Technical Field

[0001] The present invention relates to the technical field of trash cans, and in particular to an intelligent mobile trash can with both voice control and remote control functions and a control method thereof. Background Art

[0002] Trash cans (also known as waste bins or trash cans) are containers for domestic garbage. They are specially used to collect and temporarily store waste generated in daily life or public environments. They have both health protection and environmental protection functions. Most of them are made of metal (such as stainless steel) or plastic.

[0003] Traditional trash cans rely on manual lid opening or foot-operated mechanical structures, cannot respond to voice commands or remote control, have limited user experience, lack intelligent obstacle avoidance capabilities, cannot move autonomously to the user-specified location, have poor environmental adaptability, and static designs cannot cope with dynamic obstacles (such as pets and moving objects). They are prone to collisions or jams, and have insufficient detection capabilities for low obstacles (such as slippers), relying on manual intervention. Summary of the Invention

[0004] (1) Technical problems solved

[0005] In response to the shortcomings of the existing technology, the present invention provides an intelligent mobile trash can and a control method thereof with both voice control and remote control functions. It has the advantages of being more suitable for use by the elderly and children and more convenient to operate. It solves the problems of traditional trash cans relying on manual lid opening or foot-operated mechanical structures, being unable to respond to voice commands or remote control, having limited user experience, lacking intelligent obstacle avoidance capabilities, being unable to move autonomously to the user-specified location, having poor environmental adaptability, static design being unable to cope with dynamic obstacles, prone to collisions or jams, insufficient low obstacle detection capabilities, and relying on manual intervention.

[0006] Technical Solution

[0007] In order to achieve the above-mentioned purpose of being more suitable for use by the elderly and children and more convenient to operate, the present invention provides the following technical solutions: an intelligent mobile trash can with both voice control and remote control functions, comprising a trash can body, a bucket cover is installed on the top of the trash can body, a circular microphone array is installed on the front side of the trash can body, a TOF camera is installed on the front side of the trash can body, an ultrasonic sensor is installed on the outside of the trash can body, a buzzer alarm is installed on the left side of the trash can body, and a magnetic charging contact is installed on the bottom of the trash can body.

[0008] Preferably, a voice control system is installed inside the trash can body, and the voice control system includes a main control module, a voice collection module, a voice processing module, a voiceprint verification module, a dynamic navigation module and a remote control module;

[0009] The main control module controls the voice acquisition module, voice processing module, voiceprint verification module, dynamic navigation module and remote control module, so that the modules inside the trash can body work together;

[0010] The speech processing module collects speech signals through a microphone array and calculates the sound source direction through a TDOA algorithm;

[0011] The voiceprint verification module is equipped with a dual-path recurrent neural network to achieve noise separation and supports both offline instruction set and online semantic analysis modes;

[0012] The voiceprint verification module compares the pre-stored voiceprint template based on MFCC features and outputs an authorization signal;

[0013] The dynamic navigation module uses a TOF camera and ultrasonic sensor to build an obstacle map and adopts an improved A* algorithm to plan the path;

[0014] The remote control module is triggered when the voiceprint verification fails three times in a row or the environmental signal-to-noise ratio is less than 5dB, and sends a downgrade instruction through the 2.4G wireless module to activate the physical remote control to take over.

[0015] Preferably, the intelligent mobile trash can control method comprises the following steps:

[0016] Step 1: The user speaks the wake-up word, and the ring microphone array collects the voice signal and calculates the direction of the sound source using the TDOA algorithm to determine the user's location.

[0017] Step 2: A dual-path recurrent neural network model is used to perform noise reduction and feature extraction on the speech signal, and compares it with offline instruction set keywords such as open, close, and come here, or online semantic analysis such as trash can and come to me.

[0018] Step 3: If it is recognized as a valid command, go to the next step; otherwise, it will prompt “Please say it again;

[0019] Step 4: Enter the valid command into the pre-entered voiceprint template for comparison. If the verification is successful, the command will be executed. Otherwise, it will prompt that the user is unauthorized.

[0020] Step 5: The TOF camera and ultrasonic sensor scan the environment, build a real-time obstacle map, and use an improved A* algorithm to plan a path to avoid low obstacles such as pets and slippers. The mobile chassis uses Hall encoders to feedback its position, achieving centimeter-level precision movement.

[0021] Step 6: Exception handling: If the voiceprint verification fails three times in a row, or the environmental signal-to-noise ratio is less than 5dB, a downgrade command is sent via the 2.4G wireless module to activate the physical remote control to take over;

[0022] Step 7: Using Macro Design:

[0023] Long press button 1: the trash can automatically returns to the base station to charge;

[0024] Buttons 2 and 3: Control the trash can body and lid opening and closing;

[0025] Up, down, left, and right keys: control the movement of the trash can.

[0026] Preferably, the step 1, calculating the sound source orientation using the TDOA algorithm, further comprises the following steps:

[0027] Step 11: Signal acquisition: The ring microphone array receives the sound source signal and records the time domain waveform data of each channel. It is necessary to ensure that the spacing between each microphone is known.

[0028] Step 12: Delay estimation: Perform generalized cross-correlation processing on the microphone signals to calculate the delay difference between the signals;

[0029] Step 13: Determine the time difference τ by finding the peak position of the cross-correlation function.

[0030] Step 14: Calculate the azimuth angle. Calculate the angle of arrival θ of the sound source based on the TDOA value τ and the microphone spacing d, combined with the sound speed v.

[0031] Step 15: Positioning solution: Solve the nonlinear equations using Newton iteration method or least squares method to obtain the spatial position of the sound source:

[0032] Let the sound source coordinates be q = (x, y, z) and the microphone position be m i =(x i ,y i ,z i ), the speed of sound is v, and the arrival time difference is τ i1 =t i -t1

[0033]

[0034] where q LS is the least squares preliminary solution, and R1 is determined by the auxiliary variable constraints.

[0035] Preferably, if the sound source and the microphone array are located in three-dimensional space, the TDOA algorithm needs to introduce the pitch angle φ, and the formula is:

[0036]

[0037] Where d is the spacing between the microphone arrays; v is the speed of sound in the microphone array; and τ is the time difference of arrival of the signals.

[0038] Preferably, the DPRNN model performs the following specific steps on the noise reduction and feature extraction of speech signals:

[0039] Step 21: Input preprocessing, framing and overlapping: Split the speech signal into blocks of length K and 50% overlap to form a three-dimensional tensor. Normalization: Normalize the input waveform to improve training stability.

[0040] Step 22: Feature Extraction and Noise Reduction

[0041] Encoder: Use 1D convolution kernel to map the time domain signal to a high-dimensional feature space;

[0042] Step 23: DPRNN separation: Each DPRNN block contains a bidirectional LSTM / GRU, alternately processing intra-block and inter-block paths, and enhancing modeling capabilities by stacking multiple DPRNN blocks;

[0043] Step 24: Mask estimation: Generate a time-frequency mask for the mixed features to suppress the noise component.

[0044] Step 25: Signal reconstruction: Decoder: Transposed convolution maps the separated features back to the time domain waveform.

[0045] Step 26: Overlap-add: Merge the overlapping block outputs to recover the continuous speech.

[0046] Preferably, the DPRNN model loss function adopts the scale-invariant signal-to-noise ratio maximization objective:

[0047]

[0048] Among them, s target is the pure signal projection, e noise is the residual noise.

[0049] Preferably, the specific steps of step 4, outputting the authorization signal based on MFCC feature comparison with the pre-stored voiceprint template, are as follows:

[0050] Step 41: Voice collection: The user reads a fixed text 2-3 times and records the voice signal. The sampling rate is recommended to be above 16kHz and stored in mono format.

[0051] Step 42: MFCC feature extraction

[0052] Pre-emphasis: Enhance high-frequency components through a high-pass filter;

[0053] Framing and windowing: frame length 20-30 ms, frame shift 10 ms, Hamming window smoothing w(n) = 0.54-0.46cos(2πn / N));

[0054] Step 43: Mel Frequency Conversion:

[0055] Step 44: DCT coefficient calculation (MFCC extraction):

[0056]

[0057] Among them E k is the kth Mel filter energy;

[0058] Step 45: Dynamic Time Warping: Align the time axes of the test feature and the template feature and calculate the minimum cumulative distance:

[0059]

[0060] Where d(i,j) is the Euclidean distance between two frames of MFCC

[0061] Step 46: Store the MFCC feature sequence into the database;

[0062] Step 47: Set the threshold DTW distance <3.0, and generate an authorization signal if the threshold is exceeded.

[0063] Preferably, the step 5 further comprises the following steps:

[0064] Step 51: Multi-sensor Data Fusion

[0065] TOF camera: detects low dynamic obstacles through optical flow method, extracts feature points and tracks motion trajectory; Ultrasonic sensor: covers close-range blind spots, detects static obstacle outlines, and supplements camera blind spot data; Real-time obstacle map: fuses sensor data into a grid map, and marks obstacle areas as high-cost

[0066] Step 52: By improving the A* path planning, the original path is optimized using Bezier curves to reduce sharp turns:

[0067] Step 53: Generate a target speed curve based on the planned path, and adjust the motor output through PID to eliminate the accumulated error.

[0068] Beneficial effects

[0069] Compared with the prior art, the present invention provides an intelligent mobile trash can with both voice control and remote control functions and a control method thereof, which has the following beneficial effects:

[0070] 1. This intelligent mobile trash can with both voice control and remote control functions and its control method, intelligent and precise voice interaction and safety verification, sound source localization and noise reduction: A circular microphone array + TDOA algorithm achieves a sound source directional accuracy of ±5°. Combined with the DPRNN model, it maintains a command recognition rate of over 92% in noisy environments, significantly reducing the probability of false triggering.

[0071] Dual authentication: Voiceprint comparison and offline / online command analysis provide dual verification to prevent unauthorized users from manipulating the device or pets from accidentally triggering it, improving security by over 80%.

[0072] 2. This intelligent mobile trash can and its control method, which have both voice control and remote control functions, avoid obstacles through multi-sensor fusion. The TOF camera and ultrasonic sensor cover blind spots and collaboratively build a centimeter-level grid map. The improved A* algorithm introduces a dynamic steering penalty term to increase the success rate of avoiding low obstacles. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a three-dimensional schematic diagram of the present invention;

[0074] Figure 2 This is a schematic diagram of a control method for an intelligent mobile trash can according to the present invention;

[0075] Figure 3 Schematic diagram of the voiceprint comparison process of the present invention.

[0076] In the picture: 1 trash can body, 2 lid, 3 microphone array, 4 TOF camera, 5 ultrasonic sensor, 6 buzzer alarm, 7 magnetic charging contact. DETAILED DESCRIPTION

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention.

[0078] See also Figure 1-3 , an intelligent mobile trash can with both voice control and remote control functions, including a trash can body 1, a trash can cover 2 installed on the top of the trash can body 1, a circular microphone array 3 installed on the front side of the trash can body 1, a TOF camera 4 installed on the front side of the trash can body 1, an ultrasonic sensor 5 installed on the outside of the trash can body 1, a buzzer alarm 6 installed on the left side of the trash can body 1, and a magnetic charging contact 7 installed on the bottom of the trash can body 1.

[0079] Example 1

[0080] The interior of the trash can body 1 is equipped with a voice control system, which includes a main control module, a voice acquisition module, a voice processing module, a voiceprint verification module, a dynamic navigation module and a remote control module;

[0081] Among them, the main control module controls the voice collection module, voice processing module, voiceprint verification module, dynamic navigation module and remote control module, so that the modules inside the trash can body 1 work together;

[0082] The speech processing module collects speech signals through the microphone array 3 and calculates the direction of the sound source through the TDOA algorithm;

[0083] The voiceprint verification module is equipped with a dual-path recurrent neural network (DPRNN) to achieve noise separation and supports both offline instruction set and online semantic analysis modes.

[0084] The voiceprint verification module compares the pre-stored voiceprint template based on MFCC features and outputs an authorization signal;

[0085] The dynamic navigation module uses the TOF camera 4 and ultrasonic sensor 5 to build an obstacle map and uses the improved A* algorithm to plan the path;

[0086] The remote control module is triggered when voiceprint verification fails three times in a row or the environmental signal-to-noise ratio is less than 5dB. It sends a downgrade command through the 2.4G wireless module to activate the physical remote control to take over.

[0087] In this embodiment, voice control and voiceprint verification

[0088] Precise wake-up: A 3-way ring microphone array and TDOA algorithm achieve ±5° sound source localization, and the DPRNN model has a command recognition rate of >92% in noisy environments.

[0089] Double security verification: MFCC feature voiceprint comparison threshold > 0.85 combined with offline / online command analysis to prevent unauthorized operations.

[0090] Example 2

[0091] The intelligent mobile trash can control method includes the following steps:

[0092] Step 1: The user speaks the wake-up word, and the ring microphone array collects the voice signal and calculates the direction of the sound source using the TDOA Time Difference of Arrival algorithm to determine the user's location.

[0093] Step 2: Use a dual-path recurrent neural network (DPRNN) model to perform noise reduction and feature extraction on the speech signal, and compare it with offline instruction set keywords such as open, close, and come here, or online semantic analysis such as trash can and come to me.

[0094] Step 3: If it is recognized as a valid command, go to the next step; otherwise, it will prompt “Please say it again;

[0095] Step 4: Enter the valid command into the pre-entered voiceprint template for comparison. If the verification is successful, the command will be executed. Otherwise, it will prompt that the user is unauthorized.

[0096] Step 5: The TOF camera 4 and ultrasonic sensor 5 scan the environment, build a real-time obstacle map, and use an improved A* algorithm to plan a path to avoid low obstacles such as pets and slippers. The mobile chassis uses the Hall encoder to feedback its position, achieving centimeter-level precision movement.

[0097] Step 6: Exception handling: If the voiceprint verification fails three times in a row, or the environmental signal-to-noise ratio is less than 5dB, a downgrade command is sent via the 2.4G wireless module to activate the physical remote control to take over;

[0098] Step 7: Using Macro Design:

[0099] Long press button 1: the trash can 1 automatically returns to the base station to charge;

[0100] Buttons 2 and 3: Control the opening and closing of the trash can body 1 and lid 2;

[0101] Up, down, left, and right keys: control the movement of the trash can body 1 forward, backward, left, and right.

[0102] In this embodiment, degradation control: automatically switches to 2.4G remote control mode when voiceprint fails continuously or the signal-to-noise ratio is less than 5dB, and the fault recovery time is less than 1 second. Macro command shortcut operation: physical buttons support core functions such as one-button recharging and opening the cover, and the response speed is increased by 50%.

[0103] Example 3

[0104] Step 1: The TDOA algorithm calculates the sound source direction and also includes the following steps:

[0105] Step 11: Signal acquisition: The ring microphone array 3 receives the sound source signal and records the time domain waveform data of each channel. It is necessary to ensure that the distance between each microphone is known.

[0106] Step 12: Delay estimation: Perform generalized cross correlation (GCC) processing on the microphone signals to calculate the delay difference between the signals;

[0107] Step 13: Determine the time difference τ by finding the peak position of the cross-correlation function (unit: seconds).

[0108] Step 14: Calculate the azimuth angle. Based on the TDOA value τ and the microphone spacing d, combined with the sound speed v (about 343 m / s at room temperature), calculate the sound source arrival angle θ.

[0109] Step 15: Positioning solution: Solve the nonlinear equations using Newton iteration method or least squares method to obtain the spatial position x, y, z of the sound source:

[0110] Let the sound source coordinates be q = (x, y, z) and the microphone position be m i =(x i ,y i ,z i ), the speed of sound is v, and the time difference of arrival TDOA is τ i1 =t i -t1 takes the first microphone as reference

[0111]

[0112] where q LS is the least squares preliminary solution, and R1 is determined by the auxiliary variable constraints.

[0113] If the sound source and microphone array are located in three-dimensional space, the TDOA algorithm needs to introduce the pitch angle φ, and the formula is:

[0114]

[0115] Where d is the spacing between the microphone arrays; v is the speed of sound in the microphone array; and τ is the time difference of arrival of the signals.

[0116] In this embodiment, noise is suppressed by the TDOA algorithm, so that the speech recognition rate in noise is higher.

[0117] Example 4

[0118] The specific steps of the DPRNN model for noise reduction and feature extraction of speech signals are as follows:

[0119] Step 21: Input preprocessing, framing and overlapping: Split the speech signal into blocks of length K and 50% overlap to form a three-dimensional tensor of channel × block length × number of blocks. Normalization: Normalize the input waveform to improve training stability.

[0120] Step 22: Feature Extraction and Noise Reduction

[0121] Encoder: Use 1D convolution kernel such as Conv1 D to map the time domain signal to a high-dimensional feature space;

[0122] Step 23: DPRNN separation: Each DPRNN block contains a bidirectional LSTM / GRU, alternately processing intra-block and inter-block paths, and enhancing modeling capabilities by stacking multiple DPRNN blocks;

[0123] Step 24: Mask estimation: Generate a time-frequency mask for the mixed features to suppress the noise component.

[0124] Step 25: Signal reconstruction: Decoder: Transposed convolution maps the separated features back to the time domain waveform.

[0125] Step 26: Overlap-add (OLA): Merge the overlapping block outputs to recover the continuous speech.

[0126] Preferably, the DPRNN model loss function adopts the scale-invariant signal-to-noise ratio SI-SNR maximization objective:

[0127]

[0128] Among them, s target is the pure signal projection, e noise is the residual noise.

[0129] Example 5

[0130] Step 4: Specific steps for outputting the authorization signal based on MFCC feature comparison with the pre-stored voiceprint template:

[0131] Step 41: Voice acquisition: The user reads a fixed text such as a string of numbers or a specific phrase 2-3 times and records the voice signal. The sampling rate is recommended to be above 16kHz and stored in mono format.

[0132] Step 42: MFCC feature extraction

[0133] Pre-emphasis: Enhance high-frequency components through a high-pass filter y[n]=x[n]-0.97x[n-1];

[0134] Framing and windowing: frame length 20-30 ms, frame shift 10 ms, Hamming window smoothing w(n) = 0.54-0.46cos(2πn / N);

[0135] Step 43: Mel Frequency Conversion:

[0136] Step 44: DCT coefficient calculation (MFCC extraction):

[0137]

[0138] Among them E k is the kth Mel filter energy;

[0139] Step 45: Dynamic Time Warping (DTW): Align the time axes of the test feature and the template feature and calculate the minimum cumulative distance:

[0140]

[0141] Where d(i,j) is the Euclidean distance between two frames of MFCC

[0142] Step 46: Store the MFCC feature sequence into the database;

[0143] Step 47: Set the threshold DTW distance <3.0, and generate an authorization signal if the threshold is exceeded.

[0144] Example 6

[0145] Step 5 also includes the following steps:

[0146] Step 51: Multi-sensor Data Fusion

[0147] TOF camera 4: uses optical flow to detect low dynamic obstacles such as pets and slippers, extracts feature points and tracks motion trajectories; Ultrasonic sensor 5: covers a close-range blind spot of 0.1-5m, detects the outline of static obstacles, and supplements the camera's blind spot data; Real-time obstacle map: fuses sensor data into a grid map, and marks obstacle areas as high-cost obstacle layers such as costmap.

[0148] Step 52: By improving the A* path planning, the original path is optimized using Bezier curves to reduce sharp turns:

[0149] Step 53: Generate a target speed curve based on the planned path, and adjust the motor output through PID to eliminate the accumulated error.

[0150] In this embodiment, the TOF camera 0.1-4m and the ultrasonic sensor work together to build a grid map, and the improved A* algorithm introduces dynamic steering penalties to reduce the success rate of obstacle avoidance.

[0151] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0152] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent mobile trash can with both voice control and remote control functions, comprising a trash can body (1), characterized in that: The top of the trash can body (1) is installed with a bucket cover (2), the front side of the trash can body (1) is installed with a circular microphone array (3), the front side of the trash can body (1) is installed with a TOF camera (4), the outside of the trash can body (1) is installed with an ultrasonic sensor (5), the left side of the trash can body (1) is installed with a buzzer alarm (6), and the bottom of the trash can body (1) is installed with a magnetic charging contact (7).

2. The intelligent mobile trash can with both voice control and remote control functions according to claim 1, characterized in that: A voice control system is installed inside the trash can body (1), and the voice control system includes a main control module, a voice collection module, a voice processing module, a voiceprint verification module, a dynamic navigation module and a remote control module; The main control module controls the voice collection module, the voice processing module, the voiceprint verification module, the dynamic navigation module and the remote control module, so that the modules inside the trash can body (1) work in coordination; The speech processing module collects speech signals through a microphone array (2) and calculates the direction of the sound source through a TDOA algorithm; The voiceprint verification module is equipped with a dual-path recurrent neural network (DPRNN) to achieve noise separation and supports both offline instruction set and online semantic analysis modes; The voiceprint verification module compares the pre-stored voiceprint template based on MFCC features and outputs an authorization signal; The dynamic navigation module constructs an obstacle map through a TOF camera (4) and an ultrasonic sensor (5), and plans a path using an improved A* algorithm; The remote control module is triggered when the voiceprint verification fails three times in a row or the environmental signal-to-noise ratio is less than 5dB, and sends a downgrade instruction through the 2.4G wireless module to activate the physical remote control to take over.

3. A method for controlling an intelligent mobile trash can with both voice control and remote control functions, comprising the method for controlling a trash can according to any one of claims 1 to 2, characterized in that: The intelligent mobile trash can control method comprises the following steps: Step 1: The user speaks the wake-up word, and the ring microphone array collects the voice signal and calculates the direction of the sound source using the TDOA (Time Difference of Arrival) algorithm to determine the user's location. Step 2: Use a dual-path recurrent neural network (DPRNN) model to perform noise reduction and feature extraction on the speech signal, and compare it with offline instruction set keywords such as open, close, come here, or online semantic analysis such as trash can and come to me; Step 3: If it is recognized as a valid command, proceed to the next step; otherwise, it will prompt "Please say it again; Step 4: Enter the valid command into the pre-entered voiceprint template for comparison. If the verification is successful, the command will be executed. Otherwise, it will prompt that the user is unauthorized. Step 5: The TOF camera (4) and ultrasonic sensor (5) scan the environment, build a real-time obstacle map, and use the improved A* algorithm to plan the path to avoid low obstacles such as pets and slippers. The mobile chassis feedbacks the position through the Hall encoder to achieve centimeter-level precision movement; Step 6: Exception handling: If the voiceprint verification fails three times in a row, or the environmental signal-to-noise ratio is less than 5dB, a downgrade command is sent via the 2.4G wireless module to activate the physical remote control to take over; Step 7: Using Macro Design: Long press button 1: the trash can (1) automatically returns to the base station to charge; Keys 2 and 3: Control the opening and closing of the trash can body (1) and lid (2); Up, down, left, and right keys: control the movement of the trash can body (1) forward, backward, left, and right.

4. The control method of the intelligent mobile trash can with both voice control and remote control functions according to claim 1, characterized in that: The step 1, calculating the sound source direction using the TDOA algorithm, further comprises the following steps: Step 11: Signal acquisition: the ring microphone array (3) receives the sound source signal and records the time domain waveform data of each channel. It is necessary to ensure that the distance between each microphone is known. Step 12: Delay estimation: Perform generalized cross correlation (GCC) processing on the microphone signals to calculate the delay difference between the signals; Step 13: Determine the time difference τ (unit: seconds) by finding the peak position of the cross-correlation function. Step 14: Calculate the azimuth angle. Calculate the angle of arrival θ of the sound source based on the TDOA value τ and the microphone spacing d, combined with the speed of sound v (approximately 343 m / s at room temperature). Step 15: Positioning solution: Solve the nonlinear equations using Newton iteration method or least squares method to obtain the spatial position (x, y, z) of the sound source: Let the sound source coordinates be q = (x, y, z) and the microphone position be m i =(x i ,y i ,z i ), the speed of sound is v, and the time difference of arrival (TDOA) is τ i1 =t i -t1 (with the first microphone as reference) where q LS is the least squares preliminary solution, and R1 is determined by the auxiliary variable constraints.

5. The control method of the intelligent mobile trash can with both voice control and remote control functions according to claim 1, characterized in that: If the sound source and microphone array are located in three-dimensional space, the TDOA algorithm needs to introduce the pitch angle φ, and the formula is: Where d is the spacing between the microphone arrays; v is the speed of sound in the microphone array; and τ is the time difference of arrival of the signals.

6. The control method of the intelligent mobile trash can with both voice control and remote control functions according to claim 1, characterized in that: The specific steps of the DPRNN model for noise reduction and feature extraction of speech signals are as follows: Step 21: Input preprocessing, framing and overlapping: Split the speech signal into blocks of length K and 50% overlap to form a three-dimensional tensor (channel × block length × number of blocks). Normalization: Normalize the input waveform to improve training stability. Step 22: Feature Extraction and Noise Reduction Encoder: Use 1D convolution kernel (such as Conv1 D) to map the time domain signal to a high-dimensional feature space; Step 23: DPRNN separation: Each DPRNN block contains a bidirectional LSTM / GRU, alternately processing intra-block and inter-block paths, and enhancing modeling capabilities by stacking multiple DPRNN blocks; Step 24: Mask estimation: Generate a time-frequency mask for the mixed features to suppress the noise component. Step 25: Signal Reconstruction: Decoder: Transposed convolution maps the separated features back to the time domain waveform. Step 26: Overlap-Add (OLA): Merge the overlapping block outputs to recover the continuous speech.

7. The control method of the intelligent mobile trash can with both voice control and remote control functions according to claim 1, characterized in that: The DPRNN model loss function adopts the scale-invariant signal-to-noise ratio (SI-SNR) maximization objective: Among them, s target is the pure signal projection, e noise is the residual noise.

8. The control method of a smart mobile trash can with both voice control and remote control functions according to claim 1, characterized in that: The specific steps of step 4, outputting the authorization signal based on MFCC feature comparison with the pre-stored voiceprint template: Step 41: Voice collection: The user reads a fixed text (such as a string of numbers or a specific phrase) 2-3 times and records the voice signal. The sampling rate is recommended to be above 16kHz and stored in mono format. Step 42: MFCC feature extraction Pre-emphasis: Enhance high-frequency components through a high-pass filter (y[n]=x[n]-0.97x[n-1]); Framing and windowing: frame length 20-30 ms, frame shift 10 ms, Hamming window smoothing w(n) = 0.54-0.46cos(2πn / N)); Step 43: Mel Frequency Conversion: Step 44: DCT coefficient calculation (MFCC extraction): Among them E k is the kth Mel filter energy; Step 45: Dynamic Time Warping (DTW): Align the time axes of the test feature and the template feature and calculate the minimum cumulative distance: Where d(i,j) is the Euclidean distance between two frames of MFCC Step 46: Store the MFCC feature sequence into the database; Step 47: Set the threshold DTW distance <3.0, and generate an authorization signal if the threshold is exceeded.

9. The control method of the intelligent mobile trash can with both voice control and remote control functions according to claim 1, characterized in that: The step 5 further comprises the following steps: Step 51: Multi-sensor Data Fusion TOF camera (4): detects low dynamic obstacles (such as pets and slippers) through optical flow, extracts feature points and tracks motion trajectories; Ultrasonic sensor (5): covers short-range blind spots (0.1-5m), detects static obstacle outlines, and supplements camera blind spot data; Real-time obstacle map: fuses sensor data into a grid map, and marks obstacle areas as high-cost (such as the obstacle layer of Costmap) Step 52: By improving the A* path planning, the original path is optimized using Bezier curves to reduce sharp turns: Step 53: Generate a target speed curve based on the planned path, and adjust the motor output through PID to eliminate the accumulated error.

Images