An event-driven bionic vision sensor and an adaptive regulation method thereof

Abstract

The application discloses an event-driven bionic visual sensor and an adaptive regulation method thereof, and belongs to the technical field of bionic visual sensing. The sensor comprises a topological material pixel array, which is composed of M*N pixel units, each pixel unit independently detects incident light intensity and outputs a light intensity change signal; an event detection module outputs a trigger event signal when the light intensity change exceeds a preset threshold; an event frequency statistical module statistically calculates the event trigger frequency of each pixel region; a projection geometry regulation module calculates the projection angle adjustment amount θ ij of each pixel unit based on the event frequency and outputs a sensitivity adjustment instruction; an adaptive regulation module dynamically adjusts the event trigger threshold according to the instruction; and an event output interface outputs an event stream. The application realizes high-sensitivity weak light detection based on topological materials, realizes closed-loop adaptive adjustment of sensitivity through projection geometry regulation, reduces power consumption by more than 99% compared with traditional frame sensors, and is suitable for brain-like visual perception and edge intelligent computing scenes.

Description

Technical Field

[0001] This invention relates to the field of biomimetic visual sensing technology, specifically to a biomimetic visual sensor based on an event-driven mechanism, and more particularly to an adaptive event-driven visual sensor and its control method that combines a topological material pixel array with a projection geometry control module. Background Technology

[0002] Traditional solid-state image sensors (such as CCD and CMOS) employ a frame-scanning mode, acquiring complete image data at a fixed frame rate. This operating method has the following prominent problems: Severe data redundancy: When the scene is static or changes slowly, a large amount of pixel information is repeated between adjacent frames, resulting in a huge waste of storage and transmission resources; High power consumption: All pixels work continuously, performing complete analog-to-digital conversion and readout operations even if the scene remains unchanged; Limited dynamic range: Traditional frame sensors struggle to simultaneously cover both high-brightness and low-light areas, resulting in significant performance degradation in complex lighting environments; High latency: Due to frame rate limitations, capturing high-speed moving targets suffers from inter-frame latency and motion blur. In recent years, dynamic vision sensors (DVS) have solved the above problems to some extent by simulating the event-driven mechanism of the biological retina. However, existing DVS sensors still have the following shortcomings: (1) the pixel-level analog front-end circuit is complex and has low integration; (2) the sensitivity is fixed and cannot be dynamically adjusted according to the needs of the scene; (3) it lacks the ability to adaptively process the spatiotemporal distribution characteristics of events, and has a high false trigger rate in environments with strong noise or drastic changes in lighting. Summary of the Invention

[0003] Purpose of the invention This invention aims to address the shortcomings of the prior art by providing an event-driven bionic vision sensor and its adaptive control method. By combining a topological material pixel array with projection geometry control, it achieves low-power, high dynamic range event-driven visual perception with dynamically adaptive sensitivity. Technical solution (a) An event-driven bionic vision sensor An event-driven bionic vision sensor, characterized in that it comprises: 1. Topological material pixel array A two-dimensional array consisting of M×N pixel units, each pixel unit containing a photoelectric conversion structure based on a topological material. The topological material is selected from at least one of topological insulators, topological half-metals, or topological crystalline insulators, and possesses high carrier mobility and enhanced weak light response characteristics. Each pixel unit independently detects the incident light intensity and outputs a light intensity change signal. In the aforementioned topological material pixel array, each pixel unit has a topologically protected electronic transport channel, and the crosstalk suppression rate between adjacent pixels is not less than 40 dB, ensuring signal integrity under high-density integration. 2. Event Detection Module The module is electrically connected to each pixel unit of the topological material pixel array and receives the light intensity change signal output by each pixel unit. This module converts the continuous light intensity signal of each pixel into an event signal. When the light intensity change of any pixel exceeds a preset threshold, the event detection module outputs a trigger event signal corresponding to that pixel position. like Figure 2 As shown, in each pixel unit, incident light illuminates the topological material photosensitive layer. The gate electrode receives the θ_ij control signal output by the projection geometry control module to adjust the effective detection sensitivity of the photosensitive layer. The photocurrent is processed by the change detection circuit to calculate the change in light intensity ΔI. The comparator determines whether |ΔI| is ≥ threshold T. If so, an event signal is output. like Figure 3 As shown, the event generation core includes: Change detection: ΔI = I(t) - I(t - Δt) Comparator logic: |ΔI| ≥ T? like Figure 2 As shown, if |ΔI| ≥ T, then an event signal is output; otherwise, wait for the next frame to continue monitoring. The event signal includes the row address, column address, event polarity (ON event or OFF event), and timestamp information of the pixel. 3. Event Frequency Statistics Module It is connected to the event detection module and is used to count the event triggering frequency of each pixel area within a preset time window and output frequency statistics information. Specifically, the event frequency statistics module divides the M×N pixel array into several statistical sub-regions, each containing multiple adjacent pixels, and counts the total number of event triggers in each sub-region. 4. Projection geometry control module It is connected to the event frequency statistics module, receives the frequency statistics information, calculates the projection angle adjustment amount θ_ij of each pixel unit or each sub-region based on the projection geometry control algorithm, and outputs the sensitivity adjustment command. The projection geometry control algorithm includes the following steps: Map the statistical frequencies of each sub-region to a high-dimensional projection space; Based on a preset frequency distribution objective function, calculate the projection coefficient of each pixel under the current projection base; The optimal projection angle θ_ij is solved under the constraints of sensitivity adjustment and detection accuracy. Output sensitivity adjustment command. The projection geometry control module is an analog computing module or digital signal processor based on programmable topological materials. 5. Adaptive control module Connected to the projection geometry control module, it receives the sensitivity adjustment command and dynamically adjusts the event triggering threshold of the corresponding pixel unit or sub-region in the event detection module. The adaptive control module includes: Threshold storage unit: Stores the current sensitivity threshold parameters for each pixel unit or sub-region; Comparison and Update Unit: Compares the received sensitivity adjustment command with the current threshold, and gradually adjusts the threshold voltage or threshold current according to the set step size; Feedback control unit: Feeds back the updated threshold configuration to the event detection module. 6. Event Output Interface It is connected to the event detection module and outputs the trigger event signal to the external processing unit according to the preset data format. The event output interface supports standard parallel interface, MIPI CSI-2 interface or AER (Address-Event Representation) protocol output. (II) Adaptive Control Method An adaptive control method for an event-driven bionic vision sensor based on the above-mentioned sensor, characterized by comprising the following steps (such as...) Figure 4 (as shown) Step S1: Configure the projection angle θ_ij of each pixel through the projection geometry control module; Step S2: Each pixel unit monitors changes in light intensity in real time; Step S3: The event detection module determines whether the change in light intensity |ΔI| of each pixel is greater than or equal to the sensitivity threshold T. If so, generate an event signal (pixel address + polarity + timestamp); If not, continue monitoring; Step S4: The event output interface outputs the event stream in real time; Step S5: The adaptive control module calculates the event generation frequency f_ij. Step S6: Frequency Determination and Threshold Adjustment: If f_ij > f_high, increase θ_ij and decrease sensitivity; If f_ij < f_low, decrease θ_ij to increase sensitivity; Step S7: Return to step S1, update the configuration, and form a closed-loop adaptive control cycle. Wherein, the first preset threshold f_high is greater than the second preset threshold f_low. Beneficial effects Compared with the prior art, the present invention has the following beneficial effects: Based on topological material pixel array, the high carrier mobility and weak light response enhancement characteristics of topological materials are utilized to significantly improve the response speed and weak light detection capability of pixel units, with an inter-pixel crosstalk suppression rate of no less than 40 dB, and support for high-density integration. The event-driven mechanism eliminates data redundancy. The sensor only outputs event signals when the scene changes and there is no data output in static scenes. The amount of data transmitted is reduced by 80% to 90% compared with traditional frame sensors, and the power consumption is reduced by more than 99%. A projection geometry control module is introduced to adaptively and dynamically adjust the sensitivity of each pixel region based on the event frequency statistics, enabling the sensor to maintain the optimal event trigger rate under complex lighting and dynamic scenes, and avoid noise-induced false triggering and the omission of key signals. Closed-loop adaptive control forms a complete feedback loop of "event detection → frequency statistics → projection geometry control → sensitivity adjustment", realizing real-time automatic optimization of sensor sensitivity without manual intervention; It supports standardized event output interfaces, which can be directly connected to spiking neural networks (SNNs) or neuromorphic computing chips to build efficient biomimetic visual perception systems. Attached Figure Description

[0004] Figure 1 The system architecture diagram shows the overall architecture of the topology material pixel array, event detection module, event frequency statistics module, projection geometry control module, adaptive control module, and event output interface. Figure 2 A schematic diagram of the pixel unit structure and event detection principle shows the topological material of a single pixel, the photosensitive layer, the gate electrode (tuning θ), the photocurrent change detection circuit, and the comparator logic. Figure 3 System-level signal flow and core algorithm logic block diagram; event generation core (change detection ΔI, comparator) Figure 4 Flowchart of adaptive control method. Terminology Explanation

[0005] Terminology Meaning Event signal: The digital pulse signal output by the event detection module contains the following information elements: • Pixel row address and column address: Identify the spatial location where the event occurred; • Event polarity: Indicates the type of light intensity change; ON event indicates increased light intensity, and OFF event indicates decreased light intensity. • Timestamp: Marks the precise time when an event occurred. Sensitivity threshold T: The reference threshold of the comparator in the event detection module is used to determine whether the change in light intensity is sufficient to trigger an event output. • When the change in light intensity |ΔI| ≥ T, the event detection module outputs an event signal; • Otherwise, no output will be provided. The sensitivity threshold T is a function of the projection angle θ_ij, T_ij(θ). T can be continuously adjusted by adjusting θ_ij. Event frequency f_ij: Within a preset time window, the ratio of the total number of event signals generated in the (i,j)th pixel unit or the statistical sub-region containing that pixel to the width of the time window, in units of events / ms or events / s. Projection geometry control: A parameter control method based on projection geometry theory maps the performance parameters of pixel units (such as sensitivity thresholds) to a high-dimensional projection space, and achieves independent control of the working state of each pixel unit by adjusting the projection angle θ_ij. The projection angle θ_ij is the angular coordinate of the (i,j)th pixel unit in the projection space, used to characterize the control state of that pixel unit. Detailed Implementation

[0006] Example 1: Sensor Structure Example This embodiment provides an event-driven bionic vision sensor. like Figure 1 As shown, the sensor includes a topological material pixel array 100, an event detection module 200, an event frequency statistics module 300, a projection geometry control module 400, an adaptive control module 500, and an event output interface 600. The topological material pixel array 100 consists of M×N pixel units, each containing a photoelectric conversion structure based on a topological insulator Bi₂Se₃ thin film. The row and column addresses of the pixel units are encoded by row select lines and column select lines, respectively. like Figure 2 As shown, in each pixel unit, incident light illuminates the topological material photosensitive layer. The gate electrode receives the θ_ij control signal output by the projection geometry control module to adjust the effective detection sensitivity of the photosensitive layer. The photocurrent is processed by the change detection circuit to calculate the change in light intensity ΔI. The comparator determines whether |ΔI| is ≥ threshold T. If so, an event signal is output. The event detection module 200 includes an event detection circuit for each pixel unit. When the change in pixel photocurrent exceeds a threshold, it outputs a trigger event signal. The event signal includes the pixel address, polarity, and timestamp. The event frequency statistics module 300 counts the total number of events triggered in each sub-region (e.g., an 8×8 pixel block) within a 10 ms time window. The projection geometry control module 400 is an FPGA-based digital signal processor that receives the frequency statistics of each sub-region, calculates the projection angle adjustment amount θ_ij based on the preset projection basis function, and converts the result into a sensitivity adjustment command and outputs it to the adaptive control module 500. The adaptive control module 500 changes the reference voltage value of the corresponding sub-region according to the adjustment instruction, thereby adjusting the trigger threshold of the comparator in the event detection module 200. The event output interface 600 uses the AER protocol to output the triggered event in the form of a data packet. The sensor has a zero event output rate in static scenes and only outputs event data of changing pixels in motion scenes, with power consumption of 1 / 1000 of that of traditional CMOS image sensors. Example 2: Adaptive Control Method Example This embodiment provides an adaptive control method (such as...). Figure 4 As shown), based on the sensor implementation of Embodiment 1: Step S1: Configure the projection angle θ_ij of each pixel through the projection geometry control module, initialize upon power-on, and set the sensitivity threshold of all pixel units to the default value; Step S2: Each pixel unit monitors the light intensity change in real time and outputs the light intensity change signal; Step S3: The event detection module compares the change in light intensity of each pixel |ΔI| with the current sensitivity threshold T. If |ΔI|≥T, an event signal (including pixel address, polarity, and timestamp) is generated; otherwise, monitoring continues. Step S4: The event output interface outputs the event stream in real time; Step S5: The adaptive control module calculates the event generation frequency f_ij for each pixel region; Step S6: Frequency Determination and Adaptive Control If f_ij > f_high (e.g., > 1000 events / ms), it is identified as a high-activity region. Increase θ_ij and decrease sensitivity (threshold is increased by 20%). If f_ij < f_low (e.g., < 10 events / ms), it is identified as a low-activity region. Decrease θ_ij to increase sensitivity (threshold reduced by 20%). If the frequency falls between the two, the current threshold remains unchanged; Step S7: Return to step S1, update the configuration, and form a continuous closed-loop adaptive control. This control method reduces the event rate of the sensor by 60% in strong light dynamic scenes and increases the detection sensitivity by 5 times in weak light change scenes.

Claims

1. An event-driven bionic vision sensor, characterized in that, include: The topological material pixel array consists of M×N pixel units, each of which independently detects the incident light intensity and outputs a light intensity change signal. The event detection module receives the light intensity change signal and outputs a trigger event signal when the light intensity change exceeds a preset threshold. The event frequency statistics module counts the event triggering frequency of each pixel area within a preset time window; The projection geometry control module calculates the projection angle adjustment amount of each pixel unit based on the event triggering frequency and outputs a sensitivity adjustment command. The adaptive control module dynamically adjusts the preset threshold according to the sensitivity adjustment command; The event output interface outputs the trigger event signal.

2. The sensor according to claim 1, characterized in that: Each pixel unit in the topological material pixel array contains at least one topological material selected from topological insulator, topological half-metal, or topological crystal insulator.

3. The sensor according to claim 1, characterized in that: The event signal includes pixel row address, column address, event polarity, and timestamp information.

4. The sensor according to claim 1, characterized in that: The projection geometry control module is implemented based on a programmable topological material simulation calculation module or a digital signal processor.

5. The sensor according to claim 1, characterized in that: The adaptive control module includes a threshold storage unit, a comparison and update unit, and a feedback control unit.

6. The sensor according to claim 1, characterized in that: The event output interface supports AER (Address-Event Representation) protocol, MIPI CSI-2 interface, or standard parallel interface output.

7. An adaptive control method for an event-driven bionic vision sensor, characterized in that, Includes the following steps: S1: Configure the projection angle θ_ij of each pixel through the projection geometry control module; S2: Each pixel unit monitors light intensity changes in real time; S3: The event detection module determines whether the change in light intensity |ΔI| is greater than or equal to the sensitivity threshold T. If so, it generates an event signal. S4: Event output interface outputs event stream in real time; S5: The adaptive control module calculates the event generation frequency f_ij; S6: If f_ij > f_high, increase θ_ij to decrease sensitivity; if f_ij < f_low, decrease θ_ij to increase sensitivity. S7: Return to S1 to update the configuration, forming a closed-loop adaptive control.

8. The method according to claim 7, characterized in that: The first preset threshold f_high is greater than the second preset threshold f_low.

9. The method according to claim 7, characterized in that: The event signal includes pixel address, event polarity, and timestamp information.

10. An event-driven bionic vision sensor system, characterized in that, The device includes the sensor, back-end processing unit, and clock and power management unit as described in any one of claims 1 to 6, wherein the back-end processing unit receives event signals through the event output interface and performs target detection, tracking, or identification processing.

Images