Dynamic vision sensor system with static capture mode

Abstract

A dynamic vision sensing system includes a dynamic vision sensor, an AI recognition module connected to the dynamic vision sensor, and a static mode trigger module connected to the AI recognition module. The static mode trigger module is adapted to trigger intensity changes in the environment captured by the dynamic vision sensor to observe static objects of interest that are missed by conventional dynamic vision sensors. The motion recognition module is adapted to send a command to the static mode trigger module to trigger the intensity changes upon detecting motion changes in the environment. Thus, an artificial intelligence controlled electro-optical system is proposed that eliminates the blind spots of existing DVS systems by triggering effective intensity changes of static objects at the sensing end to capture objects of interest.

Description

Technical Field

[0001] This invention relates to a dynamic vision sensor (DVS) camera that utilizes AI (artificial intelligence) algorithms to detect and recognize predefined actions of objects, and in particular, eliminates blind spots in detecting static objects of interest through AI-controlled electro-optical elements. Background Technology

[0002] In recent years, DVS cameras have gained widespread attention because they can encode visual dynamic signals into asynchronous, microsecond-precision event streams, and then generate frames from these event streams to track fast-moving objects. Compared to standard frame-based cameras, DVS cameras offer significant advantages, such as a very large dynamic range, no motion blur, microsecond-level latency, and lower power consumption due to the absence of redundant information such as capturing static environmental information.

[0003] However, most DVS cameras cannot capture static objects, which are crucial in certain applications, including human tracking and SLAM (Simultaneous Localization and Mapping), where the state of slowly moving or stationary objects relative to the sensor may be lost. Therefore, existing DVS systems have blind spots in object recognition and capture. Summary of the Invention

[0004] Therefore, in one aspect, the present invention provides a dynamic vision sensing system, comprising a dynamic vision sensor, an AI recognition module connected to the dynamic vision sensor, and a static pattern triggering module connected to the AI ​​recognition module. The static pattern triggering module is adapted to trigger intensity changes in an environment captured by the dynamic vision sensor to observe static objects. The AI ​​recognition module is adapted to send a command to the static pattern triggering module to trigger intensity changes when it detects no motion changes in the environment.

[0005] In some embodiments, the AI-based motion recognition module includes an object detection module and an action classification module connected to the object detection module. The object detection module is adapted to identify objects of interest in the environment. The action classification module is adapted to detect any classified actions from the detected objects of interest.

[0006] In some embodiments, the static mode triggering module is adapted to trigger an intensity change when an action of interest is detected.

[0007] In some embodiments, the object detection module is based on deep learning and includes a bottleneck CSP (Cross Stage Partial) layer and a long short-term memory added to the bottleneck CSP layer. The bottleneck CSP layer employs additional memory elements to obtain reinforcement learning capabilities for spatiotemporal data.

[0008] In some embodiments, the action classification module is adapted to receive an input data stream from the object detection module, perform time transfer on the input data stream, and classify the actions of interest for the object of interest.

[0009] In some embodiments, if the object of interest disappears from the environment after the action classification module detects the action of interest, the action recognition module is adapted to send a command to the static mode triggering module.

[0010] In some embodiments, the static mode triggering module further includes an optical module located in the optical path before the dynamic vision sensor, and a control module connected to the optical module. The control module is adapted to output a control signal to the optical module after receiving a command from the motion recognition module, so as to trigger an intensity change.

[0011] In some embodiments, the optical module is selected from the group consisting of a liquid lens electro-optical module, a micromechanical module, and a polarization control non-mechanical module. In the case of a liquid lens electro-optical module, the module may, for example, include a liquid lens positioned at the foremost point in front of the lens module. The focusing capability of the liquid lens is periodically changed to establish small focusing movements, thereby inducing intensity changes of static objects on the dynamic vision sensor behind the lens module. Through repeated switching between defocusing and focusing, the intensity changes in the data stream captured by the DVS sensor, particularly at the edges of objects and images.

[0012] According to another aspect of the invention, a polarization-controlled optical module is provided, comprising a circular polarizer adapted to filter polarization from incoming light rays entering the circular polarizer with a predetermined chirality; and a GPA (geometric phase axis) module located after the circular polarizer in the optical path. The GPA module is adapted to change the focus of the ambient scene input to the circular polarizer when driven by a control signal.

[0013] In some embodiments, the GPA module includes a first polarization selector (PS), a first GPA, a second polarization selector, and a second GPA. The first polarization selector, the first GPA, the second polarization selector, and the second GPA are arranged sequentially in the optical path. The first polarization selector is adapted to control the chirality of the filtered polarization via a first voltage signal. The second polarization selector is adapted to control the output chirality of the first GPA via a second voltage signal. The second voltage signal is independent of the first voltage signal.

[0014] In some embodiments, the first voltage signal and the second voltage signal are configured to be inverted to change the focus of the environmental scene output by the second GPA.

[0015] In some embodiments, each of the first and second polarization selectors is an electronically controlled birefringent liquid crystal half-wave plate (LC-HWP) that can reverse the polarization state of light received on the LC-HWP.

[0016] In some embodiments, each of the first and second polarization selectors includes a ferroelectric liquid crystal half-wave plate (FLC-HWP) sandwiched between two quarter-wave plates (QWP). The FLC-HWP is connected to an electronic drive circuit.

[0017] In some embodiments, when the voltage applied to the FLC-HWP by the electronic drive circuit changes from a negative voltage to a positive voltage, the optical axis of the FLC-HWP rotates within the base plane of the FLC-HWP, or vice versa.

[0018] In some embodiments, the GPA module includes an active geometric phase axis (AGPA). The AGPA is adapted to control the focusing of an environmental scene driven by control signals.

[0019] In some implementations, the GPA module is configured to periodically change the focus of the environment scene to trigger intensity changes in static objects within the environment scene.

[0020] According to another aspect of the present invention, a method for capturing static objects using a dynamic vision sensor is provided. The method includes the steps of: triggering an intensity change in an environment captured by the dynamic vision sensor, and observing a static object in the environment.

[0021] In some embodiments, the method further includes the steps of detecting and identifying objects of interest in the environment using an AI-based algorithm, and detecting actions of interest using an action classification module. If any object of interest disappears after being identified as a classified action, a static mode is triggered, and the state and characteristics of the static object of interest are reconstructed.

[0022] In some embodiments, the object detection step further includes a convolutional neural network (CNN) backbone to extract features from the image. A bottleneck module generates a feature pyramid incorporating Long Short-Term Memory (LSTM) memory elements to detect different sizes and proportions of objects. Finally, a head module provides the final detections with labels. The sequence of detected objects of interest is fed into an action classification module. The base model uses a CNN to compute features, while a time-shifting module moves the feature map along the time dimension to capture temporal relationships for accurate action classification.

[0023] In some embodiments, the triggering step includes a microelectromechanical system (MEMS) to vibrate the sensor when the environmental scene is captured, thereby establishing the intensity change of each DVS pixel under vibrational motion.

[0024] As can be seen, the embodiments of the present invention greatly expand the functionality of the DVS camera. Not only can it track fast-moving objects as originally designed, but when it is necessary to identify and capture one or more static objects, the dynamic vision sensing system, including the DVS camera, can be switched to static capture mode. At the boundaries of static objects, it can capture intensity changes of the static objects. Therefore, the dynamic vision sensing system and method proposed in the embodiments of the present invention overcome the limitation of the DVS camera, which is only sensitive to dynamic objects.

[0025] Furthermore, with the help of artificial intelligence, especially deep learning, the dynamic visual sensing system according to some embodiments of the present invention achieves motion recognition and can identify various actions of interest (e.g., a person unconsciously falling to the ground). By training the artificial intelligence engine with more data, the number of actions that can be recognized can be increased, which opens up the possibility of applying the dynamic visual sensing system to many different industries, such as video surveillance, animal husbandry, patient monitoring, and so on. For example, the dynamic visual sensing system can be used for accident detection, such as building equipment management, shopping mall pedestrian flow detection, outdoor restricted area monitoring, environmental hazard detection and prevention.

[0026] Furthermore, since a GPA module is used in some embodiments of the present invention, static object capture can be achieved by controlling the GPA module with a voltage control signal, without involving any mechanical moving parts. Therefore, the dynamic vision sensing system with static capture mode according to these embodiments can be compact, lightweight, and consume less energy. Moreover, the dynamic vision sensing system according to embodiments of the present invention has the advantages of fast response, low power consumption, and high resolution because static capture is performed by external optics, so the high resolution of the DVS camera is not affected. Attached Figure Description

[0027] The above and further features of the present invention will become apparent from the following description of embodiments, which are provided by way of example only by comparison with the accompanying drawings, wherein:

[0028] Figure 1 This is a schematic diagram of a dynamic vision sensing system according to an embodiment of the present invention.

[0029] Figure 2 Showing Figure 1 The structure of the object detection module in the system, which is based on the improved YOLOv5 model.

[0030] Figure 3 Showing Figure 2 An exemplary structure of the bottleneck CSP LSTM layer in the object detection module.

[0031] Figure 4 Explanation Figure 1 The structure and working principle of the action classification module in the system.

[0032] Figure 5 It's a flowchart showing how to classify a disease by identifying actions of interest in an object of interest, and how to proceed with the process. Figure 1 A method for capturing static patterns in dynamic visual sensing systems.

[0033] Figure 6a The diagram illustrates the operation of a trigger optics module comprising a GPA module for a DVS sensor, according to one embodiment, when the optics module is deactivated.

[0034] Figure 6b It shows Figure 6a How the trigger optics module works when it is enabled.

[0035] Figure 7 The structure of a GPA module including a polarization selector according to one embodiment is shown.

[0036] Figure 8 The structure and operating principle of a polarization selector for a GPA module, based on an LC-HWP, are shown according to one embodiment.

[0037] Figure 9 The structure and operating principle of a polarization selector for a GPA module according to another embodiment are shown, which is based on FLC.

[0038] Figure 10 The structure of a trigger optics module according to another embodiment is shown, which includes an active GPA module.

[0039] Figure 11a and 11b Explanation Figure 10 The structure of the active GPA module and its working principle in its enabled and disabled states are described.

[0040] Figure 12a It is a description Figure 7 Finite state machine diagram of the polarization selector state of the static mode triggered electro-optic module.

[0041] Figure 12b It is a description Figure 10 Finite state machine diagram of the active GPA state of the electro-optic module triggered in static mode.

[0042] Figure 13 The structure of a trigger optical module including a MEMS (microelectromechanical system) actuator is shown according to another embodiment.

[0043] Figure 14a The structure of a trigger optics module according to another embodiment is shown, the module comprising a liquid lens.

[0044] Figure 14b Explanation Figure 14a The working principle of the trigger optics module is to capture intensity changes of static objects.

[0045] Figure 14c It is a display Figure 14a A graph showing the optical power of a liquid lens when controlled by an applied current.

[0046] Figure 14d It is a display Figure 14a A graph showing the focusing state of a liquid lens when controlled by a voltage signal input to the liquid lens.

[0047] In the accompanying drawings, similar reference numerals denote similar components in several embodiments described herein. Detailed Implementation

[0048] Figure 1 The structure of a dynamic vision sensing system according to an embodiment of the present invention is shown. The system includes a DVS camera 20 and two modules for providing static capture functionality to the DVS camera 20: an AI recognition module 34 and a static mode electro-optic triggering module 22. The DVS camera 20 is a dynamic vision sensor, as is well known to those skilled in the art. The AI ​​recognition module 34 includes two sub-modules: an action classification module 26 and an object detection module 24. The DVS camera 20 is connected to the object detection module 24, such that the output data stream of the DVS camera 20 is input to the object detection module 24. The object detection module 24 is connected to the action classification module 26 to input the output data stream from the object detection module 24 to the action classification module 26. The static mode triggering module 22 also includes two sub-modules: an electronic control module 30 and an optical module 28. The drive signal of the electronic control module 30 is regulated by the action classification module 26 and is adapted to receive command signals from the latter. The electronic control module 30 is also connected to the optical module 28 and is adapted to output control signals to drive the optical module 28 to produce a focus shift. When the command signal received by the electronic control module 30 from the motion classification module 26 becomes or remains active, the focus shift triggers an intensity change of a static object in the environment sensed at the DVS camera 20. When the motion classification module 26 of the artificial intelligence recognition module 34 detects a motion change in the environment captured by the DVS camera 20, it is adapted to send a command signal to the static mode triggering module 22 to trigger an intensity change of the static object by focusing. Changes in motion include, in particular, the actions of objects of interest in the environment, which will be described in detail below.

[0049] It should be noted that the AI ​​recognition module 34 in this embodiment, which includes the action classification module 26 and the object detection module 24, is based on artificial intelligence and is implemented through hardware, software, firmware, middleware, microcode, hardware description language, or any combination thereof. When implemented in software, firmware, middleware, or microcode, the program code or code segment that performs the necessary tasks can be stored in a machine-readable medium. One or more processors can perform the necessary tasks. On the other hand, the electronic control module 30 includes at least a combination of hardware and software, the hardware portion of which is adapted to output voltage signals as control signals to the optical module 28. The optical module 28 in this embodiment does not contain any moving parts, so it is non-mechanical. Instead, the optical module 28 contains various optical elements, which will be described in detail below, and at least one optical element is an active element that can be controlled by a voltage signal. Figure 1 The diagram illustrates the optical path indicated by arrow 32, showing how light reflected or emitted from a target (e.g., the physical environment containing the object) passes through optical module 28, reaches DVS camera 20, and is captured by it. It is optical module 28 that directly enables DVS camera 20 to capture static objects in the environment, but optical module 28 does not modify the structure of DVS camera 20. Instead, optical module 28 acts as an add-on in the optical path preceding DVS camera 20.

[0050] When the DVS camera 20 provides a sequential data stream of the sensing environment to the object detection module 24, the object detection module 24 is adapted to identify objects of interest in the environment captured by the DVS camera 20. Figure 2 A schematic diagram of the object detection module 24 is shown, which is based on the YoloV5 architecture. YoloV5 (Yolo means "You only see it once") is, as is well known to those skilled in the art, a deep learning-based object detection algorithm. Developed using the PyTorch framework, YoloV5 performs real-time object detection and recognition in images by dividing the image into a grid system. Each cell in the grid is responsible for detecting the objects within that cell. Figure 2As shown, the YOLOv5 architecture comprises three main parts: a backbone module 36, a neck module 38, and a head module 40. The backbone module 36 is a convolutional neural network that aggregates and forms image features at different granularities. The neck module 38 is a series of layers used to blend and combine image features to propagate them forward for prediction. The head module 40 consumes the features from the neck module 38 and performs box-and-class prediction steps. However, the standard YOLOv5 model has limitations for dynamic visual sensing because it is designed for frame-by-frame prediction and is not suitable for continuous video data. Therefore, the standard YOLOv5 model cannot correlate predictions from the previous frame, resulting in low prediction accuracy.

[0051] To address this deficiency, the modified YOLOv5 model used for object detection module 24 includes bottleneck CSP layers, each with an additional Long Short-Term Memory (LSTM). In other words, the bottleneck CSP LSTM layer 42 replaces the bottleneck CSP layer in the standard YOLOv5 model to connect the coroutines 44 (concats) in neck module 38 and the convolutional layers 46 in head module 40. Figure 2 As shown, this specific embodiment includes three bottleneck CSP LSTM layers 42, each connected between its respective coroutine 44 and its respective convolutional layer 46. The bottleneck CSP LSTM layers 42 enable the object detection module 24 to process continuous data correctly and efficiently; the LSTM is added in a way that maintains approximately the same processing speed and high accuracy on continuous event data. Since sudden changes in motion can affect the visible structure in a uniform data stream, the object detection module 24 provides smooth detection of such changes. The bottleneck CSP LSTM layers 42 in the object detection module 24 are designed to learn spatiotemporal features, which helps achieve robust detection on continuous event data. Figure 3 Showing Figure 2 An exemplary structure of the bottleneck CSP LSTM shows that the layer structure consists of convolutions and storage blocks.

[0052] The action classification module 26 mentioned above is suitable for detecting whether an action of interest occurs in the perceived environment. The static mode triggering module 22 is suitable for triggering an intensity change when the action classification module 26 detects an action of interest. For example, the action of interest could be an object of interest, such as a person, suddenly disappearing from the environment. Figure 4The structure of the action classification module 26 is shown, along with its operation using multiple consecutive cropped objects 48 as an example. The action classification module 26 receives an input data stream (i.e., multiple consecutive cropped objects 48 in this example) from the object detection module 24 and converts each of the multiple consecutive cropped objects 48 into a binary image 50. Then, a time-transfer module algorithm 52 is applied to the binary image 50 to classify the action of interest for the object. In doing so, the time-transfer module algorithm 52 performs time-transfer and two-dimensional convolution on the binary image 50. The time-transfer and two-dimensional convolution techniques in convolutional neural networks are well-known to those skilled in the art, so they will not be described in further detail here. After time-transfer and two-dimensional convolution, the action classification module 26 is able to determine whether the binary image 50 represents the action of interest for the object, and if so, classifies the action of interest, for example, whether a person suddenly falls to the ground.

[0053] In the introduction Figure 1 After explaining the structure and / or function of each module, what needs to be explained now is... Figure 1 The working principle of a dynamic visual sensing system, specifically how it determines whether to enter static capture mode. For example... Figure 5 As shown, the method begins at step 54, where the DVS camera 20 initially captures a video stream (i.e., continuous data). At this point, the system is detecting objects of interest in the scene, before which the system is unaware of whether any objects exist in the environment captured by the DVS camera 20. Next, the object detection module 24 attempts to detect the presence of any objects in step 56 and determines whether an object has been detected in step 58. If it is determined in step 58 that there are no moving objects in the perceived environment, the system returns to step 54 and continues detecting objects of interest in the video stream of the scene. If it is determined in step 58 that one or more objects exist in the perceived environment (e.g., a person standing there as shown in Figure 60), the method proceeds to step 62, where the object or each object is assigned a tracking ID. Once step 62 is complete, in step 64, the ROI (Region of Interest) is drawn, and the tracking IDs are displayed on, for example, a display device. Figure 5 (Not shown in the image), for the user of the dynamic visual sensing system to observe. It should be noted that step 64 is optional, and it is related to... Figure 5 The static capture method flow shown is not directly related.

[0054] After assigning a tracking ID to the object in step 62, the method also proceeds to step 66, where the object of interest is processed by the action classification module 26, and tracklets are aggregated to generate multiple consecutive objects ready for action classification. Specifically, for each tracking ID in step 62, a tracklet is created in step 66. The tracklet pooling module (not shown) maintains the coordinates of a small number of consecutive frames for a specific ID, which will be used in subsequent step 79. The object of interest can be the object detected in steps 56 and 58, or, if multiple objects are detected, then the object of interest can be one or more detected objects. Steps 66 through 74 are performed for each object of interest. After step 66, the consecutive objects are then cropped in step 68, each consecutive object being referenced... Figure 4 Cropping is performed. Then, in step 70, the action classification module 26 analyzes the binary image of the object of interest to classify actions in an attempt to identify the action of interest. The action of interest is predefined by the user of the dynamic visual sensing system and can be recognized by the action classification module 26 after being trained on a sufficient dataset. Examples of actions of interest include car accidents on highways and, as mentioned earlier, a person falling on the street. Then, the action classification module 26 makes a judgment in step 72. If it is determined in step 72 that no action of interest has been identified (e.g., normal human behavior, such as standing or sitting), then the method will not continue to activate the static capture mode, as static capture is unnecessary in this case. If it is determined in step 72 that an action of interest has occurred for the object of interest, the method proceeds to step 74, in which the action of interest is drawn on, for example, a display device for observation by the user of the dynamic visual sensing system. It should be noted that step 74 is optional and is related to... Figure 5 The static capture method flow shown is not directly related.

[0055] If an action of interest (e.g., a person falling) is determined to have occurred in step 72, the method proceeds to step 79. In step 79, the decision from step 72 and the trajectory from step 66 are merged and fed into step 78. In step 78, two conditions are checked: 1) whether an action of interest has been detected; and 2) whether the object of interest has not moved. If both conditions are true, the method proceeds to step 76, where the action classification module 26 sends a command signal to the static mode triggering module 22 based on its recognition of the action of interest. The latter switches the dynamic visual sensing system to static capture mode. In the example of a person falling on the street as the action of interest, the person disappears from the environment captured by the DVS camera 20 after falling, and the DVS camera 20 itself cannot continue to capture the person because as the person falls to the ground, he / she does not move relative to the environment, thus not producing the intensity change that the DVS camera 20 itself can capture. Therefore, static capture mode is necessary for the DVS camera 20 to capture the person falling to the ground and becoming static. Once the static capture mode is activated, the DVS camera 20, enhanced by the static mode trigger module 22, can track a static person and capture their image through focus changes. The various devices and their operation methods for the optical module 28, which assists the DVS camera 20 in capturing static objects, will be described later. Once the dynamic vision sensing system is placed in static capture mode in step 76, the object detection module 24 will periodically / continuously provide the current position of the object of interest in step 78 to see if certain conditions are met, such as the object of interest starting to move again, thus deactivating the static capture mode. It is best to minimize the time the dynamic vision sensing system is in static capture mode, because if static scenery is always captured, the primary function of the DVS camera 20—minimizing redundant information (typically, information generated by background objects does not provide useful information)—is lost. Therefore, static capture mode should only be configured when necessary, i.e., when any detected object of interest disappears from the environment due to being static at some point, for example, a person (object of interest) falls down and does not move afterward. If not necessary, the DVS camera 20 defaults to moving object tracking mode (i.e., dynamic mode). In step 78, if the condition for enabling static capture mode persists, the method returns to step 76 and repeats between steps 76 and 78.

[0056] However, if it is found in step 78 that the condition for enabling static capture mode no longer exists, the method proceeds to step 80, where static capture mode is disabled. An example of eliminating the static capture mode condition is when a person falls to the ground and quickly begins to move again, such as getting up and walking. This may indicate that the person is not injured or does not require urgent medical attention. Since the person begins to move again, static capture mode is no longer needed, but the moving object tracking mode is sufficient to continue tracking the person from that point onward.

[0057] It should be noted that step 80 does not signify the termination of the operation of the dynamic vision sensing system. On the contrary, Figure 5 The illustrated method flow runs continuously (e.g., from step 54 to step 72), and the dynamic vision sensing system continuously tracks the moving object. As soon as movement of the object of interest is detected, the dynamic vision sensing system can re-enter static capture mode.

[0058] Both the action classification module 26 and the object detection module 24 are AI-based, specifically deep learning-based. As those skilled in the art will understand, any artificial intelligence model is based on learned data. For the action classification module 26 and the object detection module 24, they are configured to recognize objects / actions after training on a large number of object images. The number of objects is infinite, and the various views of these objects are also infinite. Therefore, the training of the AI ​​model for object recognition can be refined. An object pasteaugmentation method is proposed for the object detection module. To improve the model's ability to distinguish different objects in the environment, a gallery image covering a large number of indoor objects is created. During training, a random spatial location is cropped from the gallery image and further enhanced, such as by random flipping / rotation, and then pasted onto a random location in the input image. Other standard enhancement techniques are also used.

[0059] The optical module of the static mode triggering module of the dynamic vision sensing system according to an embodiment of the present invention will now be described. For example, the optical module described below can be used as follows: Figure 1 The dynamic vision sensing system shown is an example. It should be noted that the structure and basic working principle of the optical module are not limited to any specific implementation, because any type of optical module, as long as it is conducive to the DVS camera capturing intensity changes in the environment in a controllable manner, can be used in the dynamic vision sensing system. Figure 6a and 6bAn optical module according to one embodiment of the invention is shown, which utilizes a GPA module 182. The optical module also includes a lens 188. The GPA module 182, lens 188, and DVS camera 120 are generally aligned in a straight line along the optical path (not shown) of the environment to be optically captured by the DVS camera 120. This environment is... Figures 6a-6b The environment scene 184 is represented in the diagram. For the purpose of describing the static mode triggering module, environment scene 184 is a static scene. The DVS camera 120 is an event sensor array. The lens 188 is optional; due to the role of the GPA module 182 as an add-on preceding the DVS camera 120, the lens 188 is used to refocus the system.

[0060] GPA module 182 is a non-mechanical module for polarization control. The GPA in GPA module 182, as is well known to those skilled in the art, is a specialized circularly polarization-dependent diffractive optical element with a linearly varying phase along the radial distance established by geometric phase. GPAs typically have a flat and compact appearance and provide a reconfigurable polarization optical response due to their geometric phase. Although not shown, GPAs comprise optically anisotropic materials such as liquid crystals and metasurfaces. The GPA is configured to focus light onto a DVS as a variable-axis cone lens. The transfer function immediately following the output of the optically anisotropic medium-type GPA can be expressed in matrix form as follows:

[0061]

[0062] Where T(r) is the transfer function.

[0063] It is an identity matrix.

[0064] Γ is the Pauli spin matrix, Γ is the hysteresis, r is the radial distance from the center of GPA, and α(r) is the spatial variation of the optical axis direction distribution of the anisotropic medium material. See Alwin Ming Wai TAM et al., “Bifocal optical-vortex lens with sorting of the generated non-separable spin-orbital angular-momentum states”, Phys. Rev. Appl., vol. 7, 034010 (2017), the contents of which are incorporated herein by reference.

[0065] The angular distribution of the optical axis (equivalent to the slow propagation axis in an anisotropic medium) of GPA is as follows:

[0066]

[0067] Where P is the periodicity of the optical axis direction in the GPA. In Equation (1), the complex exponent corresponding to the geometric phase component is dominated by the spatial optical axis distribution α(r) of the LC layer. The first term in Equation (2) represents the 0th-order non-diffraction component, for which the propagation behavior of light is unmodulated. The second and third terms involve the complex geometric phase exponent exp[i2α]. GPL (r)] and exp[-i2α GPL [r], which correspond to the first and -1st order diffractions of GPA, respectively. It is easy to verify that when the second term of the transfer function in equation (1) is run on a right-handed circularly polarized (RHCP) beam, i.e. When the second term disappears, the second term disappears, but when operating on a left-handed circularly polarized (LHCP) incident beam, i.e. When the output vector becomes 1, the circular polarization state of the beam reverses to LHCP. Conversely, when the matrix operator of the third term in Equation (1) is applied to an LHCP beam, the third term disappears, while when applied to an RHCP beam, the output vector becomes 1, and the circular polarization state reverses to LHCP. The opposite signs of the complex phase parameters of the second and third terms are related to the LHCP and RHCP of the incident beam, respectively, describing that the optical responses of the first and -1st order diffraction orders are different, highlighting that the GPA is circularly polarized sensitive. Therefore, manipulating the circular polarization chirality of the incident light will change the optical response, resulting in an intensity change in the sensor. Using an active liquid crystal (LC) waveplate, the circular polarization chirality of the incident light on the GPA can be controlled under electro-optic switching to establish an intensity change on the sensor when necessary for static pattern detection. Another non-mechanical method for triggering static objects in a scene is to electro-optically modulate the hysteresis (Γ) of the GPA. When the AGPA is electro-optically active and is composed of an electro-optically birefringent material, such as LC in the device, hysteresis can be manipulated, and the hysteresis becomes voltage-dependent, i.e., Γ(V). Under a given voltage, when the AGPA's delay satisfies the half-wave condition, i.e., Γ(V) = π, the first term in equation (1) corresponding to the non-diffractive wave component disappears, and since the beam polarization state is LHCP / RHCP, the AGPA behaves as an electro-optic device that converges / diversifies light, such as... Figure 10 As shown. However, when the hysteresis disappears at a given voltage, i.e., when Γ(V) = 0, the second and third terms corresponding to the diffraction terms in Equation (1) disappear, while the first non-diffraction term becomes the dominant term. This means that the beam convergence / divergence effect of AGPA is suppressed, and the device becomes isotropic. Therefore, by electro-optically adjusting the hysteresis Γ(V) from 0 and π, an intensity change of the sensor will be established to trigger static objects within the scene.

[0068] In one example, the GPA contains multiple liquid crystals (not shown) with different orientations. With these different orientations, the GPA can provide different diffraction effects to the incident light based on the circular polarization chirality of the incident light. Compared to other solutions such as electromechanical modules, the GPA is less bulky, more durable, has greater mechanical stability, and consumes less energy. Known GPA applications include optical capture, microscopy imaging through long depths of field, and material inspection (encountering material attenuation losses through the reshaping properties of non-diffractive Bessel beams). However, in this embodiment, the GPA is used for its ability to change focus in a two-dimensional (2-D) lateral direction to capture static objects by sensing intensity changes in the environment via a DVS camera. The GPA module 182 includes an electronic control module 186, which is also a power supply module, that outputs a control signal to the GPA to trigger an intensity change upon receiving a command signal from the motion recognition module.

[0069] The GPA module 182 is an active device suitable for switching between focus on and off states. A dynamic vision sensing system incorporating the GPA module 182 can operate in one of two modes, namely... Figure 6a The dynamic modes shown (e.g., for moving object tracking) and such as Figure 6b The static capture mode is shown. In dynamic mode, a constant root-mean-square (RMS) voltage signal is applied at GPA module 182, and only dynamic objects can be captured. Therefore, for Figure 6a In the static environment scene 184, the light received at the DVS camera 120 remains unchanged because the intensity curve of the focused beam at the GPA module 182 is constant and does not modulate the optical path at the DVS camera 120. Since the optical path is constant, the intensity does not change, and the static environment scene 184 cannot be captured by the DVS camera 120. As mentioned above, the DVS camera 120 only works with data streams that change in intensity. The static environment scene 184 captured in the dynamic mode of the dynamic vision sensing system appears as a hollow shape 192a without content; therefore, any object or its pose within the static environment scene 184 cannot be captured.

[0070] Figure 12a The finite state machine diagram illustrates the operational flow of an exemplary implementation of GPA module 182, which includes a polarization selector. (Refer to later...) Figure 7 Further details of this exemplary implementation are provided. Figure 7 The GPA module 182 has a first polarization selector 203 and a second polarization selector 207. Figure 6bIn the states shown, during dynamic operation mode, they are in their “low” voltage state, as described by “0 / 0” in state S1 550. The “low” voltage state is equivalent to the voltage signal being “off” if the hysteresis of the two polarization selectors 203, 207 exactly satisfies the half-wave condition without an applied voltage. Figure 12a The circular arrow d552 in state S1 550 indicates that polarization selectors 203 and 207 will always remain in state S1 550 during dynamic operation mode.

[0071] In contrast, Figure 6b In the static capture mode shown, the GPA module 182 repeatedly switches between focus "on" and focus "off", causing the focus of the incident light to change continuously. Figure 12a The finite state machine diagram summarizes the operation as follows: before triggering the static mode, polarization selectors 203 and 207 are in state S1550. When the static capture mode is activated by a specific interest action from the monitored object, each clock cycle between state S1550 and state S2 (i.e., the first polarization selector 203 is in a "low" voltage state and the second polarization selector 207 is in a "high" voltage state marked with "0 / 1") 555, the state will transition backward ("s" arrow 558) and forward ("s" arrow 556), establishing pixel movement in all directions to trigger the static capture mode. Figure 6b The intensity change in a static environment scene of 184. Figure 6b In this context, the change in focus 183 causes a change in field of view (FOV) because, as can be seen, when the focus of the GPA module 182 is turned "ON," the intersection of the principal rays in different angular domains moves backward to get closer to the DVS camera 120. Therefore, the FOV will narrow due to the angular deflection of the GPA in all directions. Through the continuous ON-OFF switching of the focus of the GPA module 182, the image received by the DVS camera 120 will therefore experience intensity variations at the boundaries of static objects in the static environment scene 184, and these boundaries can be tracked and captured by the DVS camera 120, as shown in image 192b, from which the user can easily see objects with tracked edges in the static environment scene 184. Figure 6b In the example shown, where the object is the letter "P", the two different focus states, namely focused and defocused, cause a two-dimensional shift of the pixels at the letter's boundary. Therefore, the intensity change caused by this shift can be captured by the DVS camera 120. In fact, if in Figure 6b If the static operation mode is changed to dynamic mode due to the change in trigger signal 168, then... Figure 12aThe states of polarization selectors 203 and 207 are shown to return to state S1 550, indicated by "d" arrow 559, and if the dynamic operation mode continues, they will remain in state S1, depicted by "d" loop arrow 552.

[0072] Next, Figure 7 This illustrates one possible implementation of a GPA module according to an embodiment of the present invention, which can be, for example, as... Figures 6a-6b Use it as shown. Figure 7 The GPA module 282 is based on waveplate switching and includes a circular polarizer 201, a first polarization selector 203, a first GPA 205, a second polarization selector 207, a second GPA 209, and a lens 288. All of these are arranged sequentially along the optical path, with the circular polarizer 201 positioned at the very front (i.e., closest to the environment to be captured). The second polarization selector 207 and the first polarization selector 203 can be the same or different; each is connected to its respective electronic drive circuit 211, which is part of the control module of the optical module containing the GPA module 282. The circular polarizer 201 is a passive element that allows only circularly polarized incident light with a specific chirality to pass through, in order to control the focus ultimately captured by the DVS sensor. In other words, polarizations without a predetermined chirality are filtered out from the incident light at the circular polarizer 201. The second polarization selector 207 and the first polarization selector 203 are both non-mechanical LC wave plates used to control the polarization of their respective GPAs 205 and 207, which follow the second polarization selector 207 and the first polarization selector 203. When the voltage signals provided by the first polarization selector 203 and the second polarization selector 207 are both "low", i.e. Figure 12a In state S1 550, the focus of the static object on each of the first GPA 205 and the second GPA 207 remains unchanged. The second polarization selector 207 and the first polarization selector 203 are controlled independently because they are supplied with independent voltage signals.

[0073] The phase gradients of the first GPA205 and the second GPA207 are opposite to each other to ensure that when the control signal from the electronic drive circuit 211 to the second polarization selector 207 and the first polarization selector 203 is in a "low" voltage state, for example when the command signal from the action classification module to the control module is in or remains inactive, the focusing of the entire optical system and the resulting FOV remain unchanged. When polarization selectors 203 and 207 are in a "low" voltage state... Figure 12aIn state S1 550 as described, and when a "low" voltage signal is applied to the first polarization selector 203 and the second polarization selector 207, the first GPA 205 will converge the incident beam, while the second GPA 209 will diverge the received beam, resulting in negligible focus change in the GPA module 182. Conversely, when polarization selectors 203 and 207 are in state S1 550, the second GPA 209 will converge the incident beam, while the second GPA 209 will diverge the received beam, resulting in negligible focus change in the GPA module 182. Figure 12a In state S2 555, a "low" voltage signal "0" is applied to the first polarization selector 203, while a "high" voltage signal "1" is applied to the second polarization selector 207. The first GPA 205 and the second GPA 209 simultaneously converge the received beam, resulting in an overall change in focus. In dynamic operation mode, the GPA module 182 remains in state S1 550, and the focus of the GPA module 182 is negligible; only dynamic objects are triggered. However, in static operation mode, the states of polarization selectors 203 and 207 switch between S1 550 and S2 555, causing continuous changes in overall focus and the field of view (FOV) of the environment 192b, resulting in intensity variations that can be captured by the DVS sensor even when the object is static.

[0074] The first polarization selector 203 and the second polarization selector 207 can essentially be Figure 8 The electronically controlled LC-HWP polarization selector is a half-wave plate consisting of two indium tin oxide (ITO) glass substrates 313 connected to the positive and negative terminals of the electronic drive circuit 311, with a layer 315 sandwiched between them. Each of the glass substrates 313 includes an alignment layer controlling the LC layer, oriented at 45° relative to the xy plane in the reference frame. Based on the potential difference between the positive and negative terminals, the molecular orientation of the LC 315 located between the two glass substrates 313 tilts at a specific angle towards the normal direction of the glass substrates, modifying the polarization of the light to varying degrees. Specifically, when the control signal from the electronic drive circuit 311 is "off," the polarization selector reacts like a half-wave plate, reversing the circular polarization states of the received and emitted light, such as... Figure 8 As shown in the left-hand portion. When the control signal from the electronic drive circuit 311 is "on", the polarization selector behaves as an isotropic optical element, so the polarization states of the received light and the emitted light are also opposite, as shown in the left-hand portion. Figure 8 As shown in the right-hand section. To drive the polarization selector, the control signal is preferably a bipolar AC waveform, such as... Figure 8 As shown, this ensures DC current balance at the polarization selector, thereby improving the durability of the non-mechanical polarization control module.

[0075] Next, Figure 9This illustrates another possible implementation of the polarization selector, based on a fast ferroelectric LC (FLC) waveplate switch. It is well known that FLC can provide a much faster response time (up to 1000 times) compared to traditional LC switches. Figure 8 Compared to polarization selectors in other systems, FLC-based polarization selectors offer higher voltage dynamic range and frame rate for static mode triggering in DVS systems due to their faster response time. However, it is always necessary to... Figure 9 The polarization selector in the circuit provides the voltage. Figure 9 The polarization selector comprises two QWPs 413a and 413b, with an FLC-HWP415 sandwiched between them. An electronic drive circuit 411 is connected to the FLC-HWP 415. Depending on the polarity of the applied voltage, the FLC-HWP 415 exhibits in-plane switching, with the optical axis 416 rotating 0 degrees when a negative voltage -V is applied (see...). Figure 9 (solid arrow in the image), and rotates approximately 45 degrees when a positive voltage V is applied (see solid arrow in the image). Figure 9 (The dashed arrow in the image). The optical axis of the first QWP 413a, located in front of the FLC-HWP 415, is configured to convert circularly polarized incident light into linearly polarized outgoing light, which is aligned along the x-direction of the reference frame. For the FLC-HWP 415, the optical axis is configured such that, under a certain switching polarity, the outgoing light is linearly polarized and aligned along the x-direction in the reference frame, while under the opposite polarity, the outgoing light is linearly polarized in the y-direction. Finally, the second QWP 413b converts the linearly polarized outgoing light from the FLC-HWP 415 into circularly polarized light. Depending on the driving voltage state of the FLC-HWP 415, the chirality of the circularly polarized light leaving the FLC polarization selector will remain unchanged or be opposite to that of the light entering the selector.

[0076] Figure 10 This illustrates another possible optical implementation of static mode triggering based on AGPA. For example... Figure 10 As shown, this electro-optic system includes a circular polarizer 501, a lens 588, an AGPA 515, and an electronic drive circuit 511 connected to the AGPA 515. The circular polarizer 501, AGPA 515, and lens 588 are arranged sequentially in the optical path from front to back. The circular polarizer 501 can polarize the incident light; for this purpose, the focus can be modulated by an incident signal from the electronic drive circuit 511. The AGPA 515 is an electronically controlled birefringent LC-HWP with a patterned LC arrangement structure, wherein the orientation angle of the LC molecules is linearly proportional to the radial distance from the center of the element. Figure 11a and 11bAs shown, each glass substrate 517 of the AGPA 515 includes an ITO (indium tin oxide) layer 519 and a coated photo-alignment layer 521 (which is a UV photosensitive alignment layer), which are stacked together. An LC layer 523 is placed between the two glass substrates 513. The diffraction effect of the AGPA 515 can be eliminated by controlling the delay of the component under sufficient applied voltage, as described in Equation (1) when Γ = 0.

[0077] and Figure 7-9 Compared to GPA modules based on waveboard switching, AGPA switching requires fewer components and produces higher optical throughput. However, FLC cannot be used for AGPA, thus limiting the frame rate of the DVS system.

[0078] In terms of operation, when the command signal from the motion classification module remains inactive (i.e., in dynamic mode where no object of interest is stationary), this is equivalent to a 'high' control drive voltage signal from the electronic drive circuit 511 to the AGPA 515. The diffraction of the AGPA 515 is suppressed, and the dynamic vision sensing system maintains focus. In other words, the focus of the AGPA 515 is not typically "off" when it is not powered. Therefore, during dynamic operation mode, the AGPA 515 requires a 'high' external voltage to suppress diffraction. Figure 6a The focus change in GPA module 182. Such an operation... Figure 12b The AGPA 515 is summarized from a finite state machine, and during the dynamic operating mode, it is in a state of... Figure 12b In state S1 580, "1" indicates that a voltage is applied to AGPA 515, and the loop "d" arrow 582 emphasizes that AGPA 515 is still in state S1. Figure 8 Similar to the polarization selector in the AGPA 515, the control drive voltage signal is preferably a bipolar AC waveform. In contrast, if the dynamic vision sensing system needs to operate in static capture mode, then... Figure 6a At the electronic control module 186, the common signal is valid. The control drive voltage signal from the electronic drive circuit 511 to the AGPA 515 will switch sequentially between "low" and "high" voltage states. By adjusting the hysteresis Γ in formula (1), the high and low diffraction efficiency states of the AGPA 515 are continuously modulated. Therefore, in Figure 6bIn this configuration, the focus of the GPA module 182 will be modulated in an ON-OFF manner, establishing an intensity change at DVS to capture the object of interest, even if the object is static. Section 12b outlines the operation of the static capture mode using the AGPA 515, whose state will switch back and forth between state S1 580 and state S2 (the “low” voltage state of the AGPA, denoted by “0”) 585 each clock cycle (“s” arrow 588), thus promoting an intensity change even if the object of interest is static. In fact, if the AGPA 515 satisfies the half-wave condition, i.e., Γ(V) = π in Equation (1), then the “low” control drive voltage 511 of the AGPA 515 is equivalent to the applied voltage signal being “off”, i.e., V = 0.

[0079] If the instruction signal from the action classification module is still valid, the system's focus will change periodically as AGPA 515 is continuously turned on and off.

[0080] In addition to the various GPA modules mentioned above, trigger optics modules used in dynamic vision sensing systems can be implemented using other structures. Figure 13 An electromechanical method for triggering intensity changes in static capture mode is described, based on a MEMS actuator. The MEMS actuator, upon being powered on, induces relative motion between pixels of a dynamic vision sensor and the object. Specifically, as... Figure 13 As shown, sensor 620 is placed on top of MEMS 621, which has multiple actuators 623. The actuators 623 are individually controlled by their respective drive voltages to generate vibrations in two orthogonal directions to create pixel movement functionality, thereby inducing intensity changes of a static scene on sensor 620.

[0081] Furthermore, the sensor 620 can also be vibrated via the MEMS actuator 623 to compensate for any external vibration noise affected by the environment, thereby improving the system's shock resistance stability. The MEMS actuator 623 performs vibration noise cancellation in this way.

[0082] According to another embodiment of the invention, the trigger optics module for a dynamic vision sensing system can be implemented using a liquid lens that provides periodic focusing and defocusing. For example... Figure 14a As shown, the liquid lens electro-optical module includes a liquid lens 725 positioned at the foremost point before the lens module 788. The lens focusing power of the liquid lens 725 is periodically changed to establish small focus shifts, inducing intensity changes at the DVS sensor 720 behind the lens module 788 for static scenes. Through repeated switching between defocusing and focusing, the intensity is altered in the captured data stream of the DVS sensor 720, particularly at the edges of objects and images. Figure 14bAn example using the letter "P" is shown, where a focused and out-of-focus version of the letter is combined to obtain pixel movement that tracks the boundaries of the letter and is thus captured by the DVS sensor 720. Figure 14c An example is shown showing the relationship between the magnitude of the input current to the liquid lens 725 and the variation of the lens focusing power of the liquid lens 725. Figure 14d An exemplary waveform of the voltage signal sent from the control module (not shown) to the electro-optical module of the liquid lens is displayed. Figure 14d It can be seen that when no voltage signal (i.e., "off") is transmitted to the liquid lens electro-optical module, the liquid lens is out of focus by default.

[0083] It can be seen that the DVS camera only captures the field of view (FOV) of the environment when necessary and under predetermined conditions, such as when the object recognition module and the action classification module identify the action of the object of interest. Exemplary embodiments have therefore been fully described. Although specific embodiments are mentioned in the description, it will be clear to those skilled in the art that the invention can be practiced through variations of these specific details. Therefore, the invention should not be construed as limited to the embodiments described herein.

[0084] While embodiments of the invention have been detailed and described in the accompanying drawings and foregoing description, they should be considered illustrative rather than restrictive. It should be understood that only exemplary embodiments are shown and described, and the scope of the invention is not limited in any way. It is understood that any feature described herein can be used in any embodiment. The illustrative embodiments do not exclude each other or other embodiments not mentioned herein. Therefore, the invention also provides embodiments that include combinations of one or more of the foregoing illustrative embodiments. Modifications and variations can be made to the invention without departing from its spirit and scope; therefore, only the limitations set forth in the appended claims should be applied.

[0085] The above implementation mentions the focusing of the ambient scene input to the circular polarizer, which can be altered. Note that the focal length does not change when the optical response of the GPA is modulated. Even if the state of the GPA changes, the focal length remains the same (i.e., it remains the sharpest at that focal length). More precisely, what changes is the "point spread function," which describes the distribution of focus intensity. The term "focus" is used because the scene being detected is not a single point, but rather composed of infinitesimally small points with varying intensities.

Claims

1. A dynamic visual sensing system, comprising: Dynamic vision sensor; An AI recognition module connected to the dynamic vision sensor; as well as A static pattern triggering module connected to the AI ​​recognition module; The static mode triggering module is adapted to trigger intensity changes in the environment captured by the dynamic vision sensor to observe static objects; the AI ​​recognition module is adapted to send a command to the static mode triggering module to trigger the intensity changes when it detects no motion changes in the environment.

2. The dynamic visual sensing system according to claim 1, wherein, The AI ​​recognition module further includes: Object detection module; and An action classification module connected to the object detection module; The object detection module is adapted to identify objects of interest in the environment; the action classification module is adapted to detect classified actions from the detected objects of interest.

3. The dynamic visual sensing system according to claim 2, wherein the static mode triggering module is adapted to trigger the intensity change when a specific action is classified and the object of interest cannot be captured.

4. The dynamic visual sensing system according to claim 2, wherein the object detection module is based on deep learning and includes: Bottleneck CSP layer; and Long short-term memory added to the bottleneck CSP layer; The bottleneck CSP layer employs additional memory elements to acquire reinforcement learning capabilities for spatiotemporal data.

5. The dynamic visual sensing system according to claim 2, wherein, The action classification module is adapted to receive an input data stream from the object detection module, perform time transfer on the input data stream, and classify the actions of interest of the object of interest.

6. The dynamic visual sensing system according to claim 2, wherein, If the object of interest disappears from the environment after the action classification module detects the action of interest, the AI ​​recognition module is adapted to send the command to the static mode triggering module.

7. The dynamic visual sensing system according to claim 1, wherein the static mode triggering module further comprises: An optical module is positioned in the optical path in front of the dynamic vision sensor; and The control module connected to the optical module; The control module is adapted to output a control signal to the optical module after receiving the command from the AI ​​recognition module, so as to trigger the intensity change.

8. The dynamic vision sensing system according to claim 7, wherein the optical module is selected from the group consisting of a liquid lens electro-optic mechanical module, a micromechanical module, and a polarization control non-mechanical module.

9. A polarization-controlled optical module, comprising: A circular polarizer, suitable for filtering polarization from incoming light rays that enter the circular polarizer with a predetermined chirality; The GPA module is located after the circular polarizer in the optical path; The GPA module is adapted to change the focus of the environmental scene input to the circular polarizer when driven by a control signal.

10. The polarization-controlled optical module according to claim 9, wherein the GPA module comprises: First polarization selector; First GPA; Second polarization selector; and Second GPA; The first polarization selector, the first GPA, the second polarization selector, and the second GPA are arranged sequentially in the optical path; the first polarization selector is adapted to control the chirality of the filtered polarization through a first voltage signal; the second polarization selector is adapted to control the output chirality of the first GPA through a second voltage signal; wherein the second voltage signal is independent of the first voltage signal.

11. The polarization-controlled optical module of claim 10, wherein the first voltage signal and the second voltage signal are configured to be inverted to change the focus of the ambient scene output by the second GPA.

12. The polarization control optical module according to claim 10, wherein each of the first polarization selector and the second polarization selector is an electronically controlled birefringent liquid crystal half-wave plate, capable of reversing the polarization state of light received on the birefringent liquid crystal half-wave plate without the application of voltage.

13. The polarization control optical module according to claim 10, wherein each of the first polarization selector and the second polarization selector comprises a ferroelectric liquid crystal half-wave plate sandwiched between two quarter-wave plates; the ferroelectric liquid crystal half-wave plate is connected to an electronic drive circuit.

14. The polarization control optical module according to claim 13, wherein, When the electronic drive circuit changes the applied voltage from negative to positive, the optical axis of the ferroelectric liquid crystal half-wave plate rotates substantially in a specific direction on the base plane of the ferroelectric liquid crystal half-wave plate, or vice versa.

15. The polarization control optical module of claim 9, wherein the GPA module includes an AGPA; the AGPA is adapted to control the focusing of the environmental scene driven by a control signal.

16. The polarization control optical module of claim 9, wherein the GPA module is configured to periodically change the focus of the environmental scene to trigger intensity changes of static objects in the environmental scene.

17. A method for capturing static objects using a dynamic vision sensor, comprising the following steps: The dynamic vision sensor captures changes in intensity in the environment, which are triggered by an electronically controllable optical module. as well as Static objects in the environment are observed by the dynamic vision sensor.

18. The method of claim 17, further comprising performing the following steps prior to the triggering step and the observation step: Detecting and identifying objects of interest in the environment; and Use the action classification module to detect actions of interest.

19. The method of claim 18, wherein the observation step further comprises the following steps: Receive the input data stream output by the triggering step; Features are extracted from the input data stream using a convolutional neural network; Perform time transfer on the aforementioned features; as well as The actions of interest to the object of interest are classified.

20. The method of claim 17, wherein the triggering step further comprises: The ambient light in the environmental scene captured by the dynamic vision sensor is polarized; as well as The GPA module is used as the optical module to change the focus of the environmental scene in order to observe the static object.

Images