Capacitive touch signal adaptive compensation method and device

Abstract

The application relates to the technical field of capacitive touch control, and discloses a capacitive touch signal adaptive compensation method and device. The method comprises the following steps: collecting original capacitive signal grid data in an initial contact stage, and constructing a dynamic contact quadtree; based on the quadtree, running a first point position prediction algorithm for contact dynamics sensing, and constructing a stable coordinate prediction mapping with confidence; based on the mapping, performing multimodal first point decision and touch event generation, outputting a touch event and feeding back to an operating system. The application captures initial contact signal characteristics through a dynamic contact quadtree, combines a stable coordinate prediction mapping and multimodal decision, solves the problem of jumping coordinates in the initial touch stage of an existing system, improves touch accuracy in the scene of a vehicle central control, an industrial HMI and the like, reduces operation risk, and expands the application value of capacitive touch control technology.

Description

Technical Field

[0001] This invention relates to the field of capacitive touch technology, and discloses a capacitive touch signal adaptive compensation method and device. Background Technology

[0002] Capacitive touch technology is widely used in various touch-interactive devices. Its core principle is to establish an electric field through an electrode network on the touch panel. When a user's finger, a gloved finger, or a passive touch tool approaches or touches the panel, it changes the coupling capacitance between the electrodes. The touch system detects this capacitance change and uses algorithms to analyze and determine the touch position and action, thus achieving human-computer interaction. This technology features high responsiveness, long lifespan, and good scratch resistance. It supports multi-touch and complex gesture recognition, and is suitable for various application scenarios such as smartphones, tablets, in-vehicle central control systems, industrial human-machine interfaces, and medical equipment. To ensure reliability, related systems typically possess a certain degree of anti-interference capability and environmental adaptability, enabling basic touch position recognition and output under different usage conditions.

[0003] However, existing capacitive touch systems still have shortcomings in practical applications, namely insufficient accuracy in coordinate output during the initial touch phase. This problem arises because when a user's touch tool first touches the screen and has not yet formed a stable contact, the contact area is small, resulting in a weak coupling capacitance signal. The touch system often cannot immediately output accurate coordinates, instead producing 1-3 frames of fluctuating coordinates, making it difficult for the touch system to quickly acquire stable feature data. Although subsequent trajectories can be smoothed through filtering, the initial coordinates of the first effective touch still have significant errors. In high-reliability scenarios requiring precise touch control, such as in-vehicle central control systems, industrial HMIs, and medical equipment, this can easily lead to accidental clicks on adjacent buttons, misinterpretations of clicks as swipes, or triggering incorrect commands, interfering with the normal operation of the device. Solving this problem can improve the accuracy of touch interaction in high-reliability scenarios, reduce operational risks caused by initial coordinate deviations, and further expand the application value of capacitive touch technology. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, the present invention provides a capacitive touch signal adaptive compensation method and apparatus, comprising:

[0005] An adaptive compensation method for capacitive touch signals includes the following steps:

[0006] Step 1: Collect the original capacitance signal grid data during the initial contact stage and construct a dynamic contact quadtree; Step 2: Based on the dynamic contact quadtree, run the contact dynamics sensing first point position prediction algorithm to construct a stable coordinate prediction mapping with confidence.

[0007] Step 3: Based on the stable coordinate prediction mapping, perform multimodal first point decision and touch event generation, output touch events that conform to human-computer interaction semantics and feed them back to the operating system.

[0008] Furthermore, when constructing the dynamic contact quadtree, it is necessary to first determine the non-zero response region based on the two-dimensional matrix of initial contact capacitance change. The non-zero response region refers to the continuous area formed on the touch screen by the sensing units corresponding to the elements in the two-dimensional matrix of initial contact capacitance change whose values ​​are greater than the static noise threshold. While performing recursive quadtree partitioning on each non-zero response region, each node of the dynamic contact quadtree is assigned geometric boundaries, signal strength, time derivative, spatial gradient vector, and contact morphology label attributes.

[0009] Furthermore, the structural features of the dynamic contact quadtree include a root node, child nodes, and leaf nodes; the root node corresponds to the initially determined single non-zero response region; the child node corresponds to the sub-region generated during the recursive quadtree partitioning process; and the leaf node corresponds to the smallest sub-region formed after the recursive quadtree partitioning terminates.

[0010] Furthermore, the row dimension of the two-dimensional matrix of initial contact capacitance change corresponds to the row electrode number of the touch sensor, and the column dimension corresponds to the column electrode number of the touch sensor; each element in the two-dimensional matrix of initial contact capacitance change is the average capacitance change of the corresponding sensing unit in the first N sampling periods.

[0011] Furthermore, the recursive quadtree partitioning method is as follows: set the non-zero response region as the current region to be partitioned, and determine the geometric boundary of the region; calculate the signal variance of the current region to be partitioned; if the signal variance exceeds the dynamic noise threshold, partition the region into four sub-regions of equal area; if the signal variance is less than or equal to the dynamic noise threshold, stop partitioning the region and use the region as a leaf node of the dynamic contact quadtree; repeat the above operation for each sub-region until the signal variance of all sub-regions is less than or equal to the dynamic noise threshold, and stop the recursive partitioning.

[0012] Furthermore, in the contact dynamics sensing first point position prediction algorithm, the adaptive similarity weighted regression step is as follows: construct a similarity function, calculate the kernel response value based on the lowest common ancestor depth between leaf nodes and the node signal distance; combine the sum of the kernel response values ​​of each leaf node with the signal strength to calculate the weight of the leaf node and normalize it; and sum the predicted coordinates based on the geometric boundary center coordinates of the leaf node and the corresponding weights.

[0013] Furthermore, the method for calculating the node signal distance is as follows: select any two leaf nodes from the dynamic contact quadtree as the calculation objects; obtain the geometric boundary center coordinates of the two leaf nodes and calculate the Euclidean distance between the coordinates; obtain the signal strength of the two leaf nodes and calculate the absolute difference of the signal strength; perform weighted correction on the Euclidean distance and the absolute difference of the signal strength, introduce a distance correction coefficient determined based on the difference in signal variance of the non-zero response regions to which the two nodes belong, and finally obtain the node signal distance.

[0014] Furthermore, the method for constructing the stable coordinate prediction mapping is as follows: taking the leaf node parameters of the dynamic contact quadtree as input, the leaf node parameters include the geometric boundary center coordinates, signal strength, time derivative, and spatial gradient vector; substituting the leaf node parameters into the contact dynamics sensing first point position prediction algorithm to calculate the predicted coordinates; obtaining the confidence value based on the prediction variance and prior factors; the mapping relationship of the stable coordinate prediction mapping is to input the leaf node parameters of the dynamic contact quadtree and output the corresponding predicted coordinates and confidence.

[0015] Furthermore, the method for multimodal first point decision is as follows: if the confidence level of the stable coordinate prediction mapping is not lower than a preset threshold, a high-confidence click event is directly triggered; if the confidence level is lower than the preset threshold but the original capacitance signal continues to increase, a delayed confirmation mode is started and the signal change is continuously monitored; if the contact shape label is dot-shaped and the confidence level is at a medium level, a passive pen enhancement mode is enabled to improve positioning accuracy.

[0016] A capacitive touch signal adaptive compensation device, characterized in that the device stores a computer program adapted for loading by a processor to execute a capacitive touch signal adaptive compensation method as described above, the device comprising:

[0017] Data acquisition and quadtree construction unit: used to acquire raw capacitance signal grid data during the initial contact phase and construct a dynamic contact quadtree;

[0018] Mapping construction unit: used to run the contact dynamics-aware first point position prediction algorithm based on the dynamic contact quadtree to construct a stable coordinate prediction mapping with confidence.

[0019] Decision and event generation unit: Based on the stable coordinate prediction mapping, it performs multimodal first point decision and touch event generation, outputs touch events that conform to human-computer interaction semantics and feeds them back to the operating system.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention constructs a dynamic contact quadtree by collecting the original capacitance signal grid data in the initial contact stage, constructs a stable coordinate prediction mapping by combining the first point position prediction algorithm of contact dynamics perception, and performs multimodal first point decision based on the mapping, which accurately solves the problem of "jump coordinates" caused by small contact area and weak coupling capacitance signal in the initial touch stage of the existing capacitive touch system, and eliminates the initial coordinate error of the first effective touch.

[0021] The dynamic contact quadtree can fully capture the evolution characteristics of the initial contact signal from weak to strong and the contact area from small to large. The stable coordinate prediction mapping quantifies the coordinate reliability through confidence measurement, and the multimodal decision flexibly outputs effective touch events according to the scenario. The three work together to ensure that high-reliability scenarios such as vehicle central control, industrial HMI, and medical equipment achieve accurate touch, avoid the risks of accidentally clicking adjacent buttons, misjudging clicks and swipes, and ensure the normal operation of the equipment.

[0022] Meanwhile, this invention requires no additional hardware support and achieves adaptive compensation solely through algorithm optimization. While controlling costs, it is compatible with various touch methods such as fingers, gloved fingers, and passive touch tools, as well as different environmental conditions. This further enhances the environmental adaptability and scenario adaptability of capacitive touch systems and significantly expands the application value of capacitive touch technology. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of an adaptive compensation method for capacitive touch signals according to the present invention;

[0025] Figure 2 A flowchart illustrating the construction of a dynamic contact quadtree in an embodiment of the present invention;

[0026] Figure 3 This is a flowchart of multimodal first-point decision and touch event generation in an embodiment of the present invention;

[0027] Figure 4 This is a functional block diagram of a capacitive touch signal adaptive compensation device according to the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Example 1:

[0030] Please see Figure 1 As shown, this embodiment provides an adaptive compensation method for capacitive touch signals, including:

[0031] Step 1: Collect raw capacitance signal grid data during the initial contact phase and construct a dynamic contact quadtree;

[0032] First, the raw capacitance signal grid data is acquired. This raw capacitance signal grid data refers to the capacitance change data collected by the touch sensor at the intersections of the row and column electrode grids during the initial contact phase. The initial contact phase refers to the period from when a user's finger, gloved finger, passive touch tool, or non-conductive touch tool first touches the touchscreen surface until the contact stabilizes. This period is defined by the touch sensor's sampling frequency, monitored by observing the capacitance signal rising from a no-contact reference value until the signal change rate falls below a preset stabilization threshold, ensuring a complete transition from small to large contact area and from weak to strong coupling capacitance. The row and column electrode grid of the touch sensor is a matrix sensing structure formed by the intersection of several parallel row electrodes and several parallel column electrodes. The density of the row and column electrodes is set according to the touchscreen's resolution requirements. Each intersection point corresponds to a sensing unit on the touchscreen, and the capacitance change of each sensing unit directly reflects the touch interaction state at that location.

[0033] During the data acquisition process, the capacitance change of each sensing unit (denoted as ΔC) needs to be calculated, and ΔC is used as the core parameter of the original capacitance signal grid data. The calculation method for ΔC is as follows: subtract the reference capacitance value of a sensing unit in a non-contact state from the real-time capacitance value of that sensing unit within the current sampling period. The reference capacitance value is obtained through a calibration process of the touch system after each startup without any touch operation. Specifically, the capacitance values ​​are continuously collected for several non-contact sampling periods, and the average value is taken as the reference capacitance value of the sensing unit to eliminate the influence of the sensor's inherent capacitance and environmental parasitic capacitance on the ΔC calculation. Meanwhile, the collected raw capacitance signal grid data needs to cover the sliding window consisting of the most recent N sampling periods in real time. The value of N is determined by calculating the estimated duration of the initial contact phase and the sampling frequency of the touch sensor to ensure that the data in the sliding window can fully reflect the signal evolution process of the initial contact phase and avoid the loss of key signal features due to insufficient number of periods in the window. When the initial contact phase starts, the sliding window gradually accumulates data from the first sampling period until the window contains N periods of data. For each additional period of data, the earliest period of data is removed according to the rules.

[0034] After acquiring the raw capacitance signal grid data, the raw capacitance signal grid data is organized into a two-dimensional matrix of initial contact capacitance change based on spatial location. The row dimension of this matrix corresponds to the row electrode number of the touch sensor, and the column dimension corresponds to the column electrode number. Each element in the matrix is ​​the mean ΔC value of the sensing unit at the intersection of the corresponding row and column electrodes over the first N sampling periods. If the ΔC of a sensing unit in a certain sampling period is less than the static noise threshold, then ΔC in that period is determined to be an invalid noise signal, and its value is set to 0 before being included in the mean calculation. If ΔC is greater than or equal to the static noise threshold, the original ΔC value is retained for the mean calculation. The static noise threshold refers to the upper limit of the normal fluctuation range of the sensing unit capacitance value in the non-contact state of the touch sensor. It is calculated by statistically analyzing the standard deviation of ΔC over several sampling periods in the non-contact state and multiplying it by a preset confidence coefficient. This confidence coefficient is determined based on the anti-interference requirements of the touch system and is usually based on statistical theory to ensure the probability that normal noise is covered.

[0035] Subsequently, based on the two-dimensional matrix of initial contact capacitance change, non-zero response regions are determined. These non-zero response regions refer to the continuous areas on the touchscreen formed by the sensing units corresponding to elements in the two-dimensional matrix of initial contact capacitance change whose element values ​​(i.e., the mean ΔC) are greater than the static noise threshold. If multiple discontinuous regions exist, each discontinuous region is treated as an independent region to be partitioned, and subsequent quadratic partitioning is performed on each. The purpose of determining non-zero response regions is to focus on the core signal region of the initial contact, avoid invalid calculations for blank areas with no touch response, and reduce the computational complexity of subsequent processing.

[0036] Next, a recursive quadtree partitioning is performed on each non-zero response region to gradually construct a dynamic contact quadtree. Please refer to the detailed process below. Figure 2 As shown, the recursive quadrilateral partitioning method is as follows:

[0037] The first step is to take a single non-zero response region as the current region to be segmented and determine the geometric boundary of the region. The geometric boundary is defined by the range of row electrode numbers and column electrode numbers of all sensing units in the region, forming a rectangular bounding box.

[0038] The second step is to calculate the signal variance of the current region to be segmented. The signal variance is the statistical variance of the mean ΔC of all sensing units in the region, which is used to characterize the uniformity of the signal in the region. The larger the signal variance, the more significant the difference in touch coupling strength at different locations in the region, and further segmentation is needed to capture local signal features.

[0039] The third step is to compare the signal variance of the current region to be partitioned with the dynamic noise threshold. If the signal variance is less than or equal to the dynamic noise threshold, the partitioning of the region is stopped and the region is used as a leaf node of the dynamic contact quadtree. If the signal variance is greater than the dynamic noise threshold, quadtree partitioning is performed on the region, that is, the rectangular bounding box of the current region to be partitioned is divided into four sub-regions of equal area along the row and column directions, and each sub-region is used as a new region to be partitioned.

[0040] The fourth step is to repeat the operations of the second and third steps for each new region to be subdivided until all regions satisfy the condition that the signal variance is less than or equal to the dynamic noise threshold, and then stop the recursive subdivision.

[0041] The dynamic noise threshold is a threshold squared value that is dynamically adjusted based on the signal strength of the current non-zero response area, used to determine whether the area needs further subdivision. The signal strength is defined as the mean ΔC. The dynamic noise threshold is calculated as follows: first, obtain the mean ΔC of the current area to be subdivided, multiply this mean by the dynamic adjustment coefficient, and then add it to the static noise threshold to obtain the dynamic noise threshold; then, square the dynamic noise threshold to obtain the squared dynamic noise threshold. The dynamic adjustment coefficient is determined based on the preliminary judgment result of the touch tool type. For example, when the preliminary judgment is a passive pen or a non-conductive touch tool, the dynamic adjustment coefficient is smaller based on the area size of the non-zero response area and the overall level of the mean ΔC, in order to achieve more refined subdivision and capture weak and concentrated signal features; when the preliminary judgment is a finger, the dynamic adjustment coefficient is larger to avoid excessive subdivision and increased computation. The preliminary judgment is made by detecting differences in signal characteristics such as the size of the touch area and the change in capacitance (mean ΔC). Passive pens and similar devices typically have small areas and weak signals, while fingers have large areas and strong signals.

[0042] During the recursive quadtree partitioning process, attribute information is simultaneously assigned to each node of the dynamically contacting quadtree, forming a complete tree structure. The structural characteristics of the dynamically contacting quadtree are as follows:

[0043] (1) Root node: Corresponds to the initially determined single non-zero response region. Its attribute information includes the root node geometric boundary, root node signal strength, root node time derivative, root node spatial gradient vector, and root node contact morphology label.

[0044] (2) Child node: The corresponding sub-region generated during the recursive subdivision process. Its attribute information is consistent with the attribute information type of the root node, but the calculation range is limited to the sensing unit within the corresponding sub-region. The contact morphology label of the child node is also initially marked as undetermined and will be updated after the subdivision of its next level sub-region is completed.

[0045] (3) Leaf node: The smallest sub-region formed after the recursive partitioning terminates. Its attribute information includes geometric boundary, signal strength, time derivative, spatial gradient vector, and contact morphology label. The contact morphology label is determined as follows: the ratio of the distribution range of the mean ΔC in the sub-region corresponding to the leaf node to the area of ​​the sub-region is calculated. If the ratio is greater than the morphology determination threshold, it is marked as point, indicating signal concentration, corresponding to the contact characteristics of a passive pen or non-conductive touch tool. If the ratio is less than the morphology determination threshold and the area of ​​the sub-region is greater than the area determination threshold, it is marked as surface, indicating uniform signal distribution, corresponding to the characteristics after the finger contact area expands. If the ratio is less than the morphology determination threshold but the area of ​​the sub-region is less than or equal to the area determination threshold, and the sub-region is located at the edge of the non-zero response area, it is marked as edge-dominant, indicating signal distribution along the edge, corresponding to the edge effect of finger or glove contact. The morphology determination threshold and the area determination threshold are both determined by statistical analysis of a large number of sample data from different touch tools to ensure accurate differentiation of different contact morphologies.

[0046] Furthermore, the "dynamic" characteristic of the dynamic contact quadtree is reflected in two aspects: First, as the sampling period of the initial contact stage progresses, for each additional sampling period of the original capacitance signal grid data, the signal strength, time derivative, and spatial gradient vector of each node in the dynamic contact quadtree need to be updated to ensure that the node attributes can reflect the evolution trend of the signal in real time; Second, if the new data causes a significant change in the range of the non-zero response region, such as the expansion of the contact area leading to the expansion of the non-zero response region, then a recursive quadtree partitioning needs to be performed on the newly added non-zero response region, and the newly generated leaf nodes need to be connected to the corresponding level of the dynamic contact quadtree to ensure that the tree structure can cover the complete contact area.

[0047] The effects of step one are mainly reflected in the following aspects:

[0048] Traditional methods treat the capacitance signal in the initial contact phase as isolated sensing unit data, failing to capture the continuous expansion of the contact area from point to surface, and unable to distinguish the contact dynamics patterns of different touch tools, such as the area expansion of a finger versus the dotted hold of a passive pen. This step, however, constructs a two-dimensional matrix of initial contact capacitance changes, transforming discrete ΔC data into a spatially correlated overall data structure. Then, through recursive decomposition of a dynamic contact quadtree, the spatial hierarchy of the signal is preserved, enabling the system to perceive the signal distribution characteristics of the contact area comprehensively from global to local, laying the foundation for subsequent differentiation of different touch scenarios. This step only decomposes non-zero response regions, and the decomposition termination condition is controlled by a dynamic noise threshold. The final number of leaf nodes is typically only related to the signal complexity of the contact area, and the total number of nodes is much smaller than the total number of sensing units in the touch sensor. All calculations can be performed using fixed-point integers, requiring no floating-point arithmetic unit support. This design avoids the high memory and computing resource consumption of full-mesh filtering schemes, ensuring rapid execution on embedded platforms such as automotive MCUs and industrial HMIs. Both execution time and memory usage are kept within acceptable system limits, balancing the integrity and real-time performance of signal representation. The differential geometric features embedded in node attributes provide structured priors for subsequent predictions. Each node in the dynamic contact quadtree not only contains geometric location and signal strength but also incorporates the time derivative (reflecting the rate of signal change) and spatial gradient vector (reflecting the spatial distribution trend of the signal). These differential geometric features effectively characterize the evolution trajectory of the signal during the initial contact phase. For example, when a finger touches the screen, the time derivative is positive and gradually decreases (signal growth slows down), and the spatial gradient vector radiates (area expansion). When a passive pen touches the screen, the time derivative is close to zero (no significant signal growth), and the spatial gradient vector is concentrated (signal is fixed). These features allow subsequent steps to directly distinguish different touch scenarios based on the tree structure without additional scene identification information, solving the problem that traditional prediction models cannot adapt to multi-touch tools. Adaptive adjustment of the dynamic noise threshold improves robustness in complex environments. Traditional fixed-threshold segmentation schemes are prone to over-segmentation (misclassifying noise as signal features) or under-segmentation (missing weak signal features) in low-temperature or high-noise environments. This step adjusts the segmentation accuracy by combining a dynamic noise threshold with the initial judgment results of the touch tool. In low-temperature environments (where signals are weak and noise is relatively significant), the dynamic noise threshold can be reduced to achieve finer segmentation and capture weak signals; in normal environments, the dynamic noise threshold can be increased to avoid invalid calculations, ensuring that a dynamic contact quadtree that accurately reflects the contact state can be generated in different environments, thus guaranteeing the accuracy of subsequent first-point position prediction.

[0049] Step 2: Based on the dynamic contact quadtree, construct a stable coordinate prediction mapping with confidence using the contact dynamics-aware first point position prediction algorithm;

[0050] First, the dynamic contact quadtree is mapped to a locally Lipshitz continuous mapping space. This locally Lipshitz continuous mapping space refers to a set of functions defined in a discrete space that satisfies the local Lipshitz continuity condition. Its core function is to transform the tree structure and node attributes of the dynamic contact quadtree into a mathematical space usable for numerical computation, ensuring that the output of the subsequent prediction algorithm remains continuous as the input (node ​​attribute changes) changes, avoiding abrupt changes in prediction coordinates due to local signal fluctuations. Specifically, the mapping process is as follows: each leaf node of the dynamic contact quadtree is used as a primitive of the mapping space; the geometric boundary center coordinates of the leaf node are used as the spatial coordinates of the primitive; and the signal strength, time derivative, and spatial gradient vector of the leaf node are used as feature parameters of the primitive. After standardization of the feature parameters, the basis vectors of the mapping space are constructed. The standardization of the feature parameters involves dividing each feature parameter by the sum of a standardized baseline value calculated based on the exponential sliding maximum and a minimum positive number ε, ensuring that the standardized feature parameters have a uniform value range and eliminating the influence of differences in the magnitude of different feature parameters on the mapping space. The exponential sliding maximum value is obtained by fusing the historical sliding maximum value with the maximum value among all current leaf nodes. Specifically, the calculation method is: Exponential sliding maximum value = α × maximum value among all current leaf nodes + (1-α) × historical sliding maximum value, where α is a smoothing coefficient, determined through touch system calibration. The calibration process aims to ensure that "when a new leaf node exhibits an extreme value, the standardized baseline value changes only smoothly without abrupt changes," iteratively adjusting α until the local Lipshitz continuity requirement is met. The initial value of the historical sliding maximum value is the maximum value among all leaf nodes in the first frame after the touch system starts. In subsequent frame updates, the exponential sliding maximum value of the previous frame is retained as the historical sliding maximum value of the current frame. ε is a preset, extremely small positive number, such as 1e-6, used to avoid division by zero errors when the standardized baseline value is 0, ensuring the stability of numerical calculations. Through this mapping process, the tree structure of the dynamic contact quadtree is transformed into a set of primitives in the mapping space. The position and characteristics of each primitive correspond to the spatial distribution and evolution characteristics of the original contact signal, providing structured mathematical input for subsequent prediction algorithms.

[0051] Subsequently, the initial position prediction algorithm for contact dynamics sensing is run. The core of this algorithm is to perform adaptive similarity-weighted regression in a local Lipshitz continuous mapping space. This adaptive similarity-weighted regression refers to a regression method that constructs a similarity function to weight the contribution of each primitive (i.e., the leaf nodes of the dynamic contact quadtree) in the mapping space, ultimately obtaining the predicted coordinates. Its adaptive characteristic is reflected in the fact that the weights of the similarity function are dynamically adjusted according to the spatial correlation and signal evolution similarity of the leaf nodes to adapt to the contact dynamics characteristics under different touch scenarios. Specifically, the execution steps of the adaptive similarity-weighted regression are as follows:

[0052] The first step is to construct a similarity function and calculate the kernel response value. The similarity function, denoted as K(p,q), is used to measure the correlation between two leaf nodes in the mapping space, where p and q represent any two leaf nodes in the dynamic contact quadtree. The value of the similarity function K(p,q) is determined by a weighted average of the lowest common ancestor depth (LCA) of leaf nodes p and q, and the node signal distance between them. Specifically, the LCA depth reflects the spatial hierarchical correlation between the two nodes, while the node signal distance reflects the signal evolution similarity. The combination of these two factors provides a comprehensive characterization of node correlation.

[0053] Specifically, the method for calculating the lowest common ancestor (LCA) depth of leaf node p and leaf node q is as follows: First, traverse the hierarchical structure of the dynamic contact quadtree to determine all ancestor nodes of leaf node p (all nodes from leaf node p upwards to the root node) and all ancestor nodes of leaf node q; then, select the common ancestor node from the set of ancestor nodes of the two nodes, and define the common ancestor node that is farthest from the root node (i.e., at the highest level) as the LCA node; finally, calculate the path length from the LCA node to the root node, which is the LCA depth. The range of the LCA depth is consistent with the maximum depth of the dynamic contact quadtree. The larger the depth value, the closer the spatial positions of the two nodes are, and the stronger the macro-region consistency; the smaller the depth value, the more dispersed the spatial positions of the two nodes are, and the weaker the macro-region consistency.

[0054] Specifically, the method for calculating the node signal distance between leaf node p and leaf node q is as follows: The node signal distance is used to measure the similarity of the signal evolution trajectories of two nodes, and its calculation is based on two differential geometric features of the leaf node: the time derivative and the spatial gradient vector. First, the absolute difference between the time derivatives of leaf node p and leaf node q is calculated, denoted as Δd. Second, the cosine of the angle between the spatial gradient vectors of leaf node p and leaf node q is calculated, denoted as cosθ. The closer this value is to 1, the more consistent the spatial gradient directions of the two nodes are, and the more similar the signal spatial distribution trends are. Finally, Δd and (1-cosθ) are weighted and summed to obtain the node signal distance. The weight coefficients of the weighted summation are determined through training with a large amount of sample data from different touch scenarios to ensure that the contribution values ​​of the time derivative difference and gradient direction difference to the node signal distance conform to the actual contact dynamics. For example, in the passive pen touch scenario, the weight coefficient of the time derivative difference will be appropriately increased because there is no significant signal growth when the passive pen is in contact, and the stability of the time derivative has a more significant impact on the evolutionary similarity.

[0055] After obtaining the lowest common ancestor depth and node signal distance, the similarity function K(p,q), i.e., the kernel response value, is calculated using a weighted formula. In this weighted formula, the weight coefficients for the lowest common ancestor depth and node signal distance need to be dynamically adjusted according to the touch scenario. For example, in low-temperature environments, the weight coefficient for node signal distance will be appropriately increased because low temperatures lead to enhanced nonlinearity in signal evolution trajectories, making the impact of evolutionary similarity on node correlation more prominent. All leaf nodes in the dynamic contact quadtree are paired, and the kernel response value for each pair of leaf nodes is calculated, resulting in a set of kernel response values. This set is used for subsequent calculation of leaf node weights.

[0056] The second step is to calculate the leaf node weights. The leaf node weight refers to the contribution of each leaf node to the predicted coordinates. Its value is determined by the kernel response values ​​of the leaf node and all other leaf nodes, as well as the signal strength of the leaf node itself. The specific calculation method is as follows: First, for each leaf node p, extract all kernel response values ​​containing leaf node p from the kernel response value set, i.e., the kernel response values ​​of leaf node p and all other leaf nodes, and sum these kernel response values ​​to obtain the total kernel response of leaf node p; second, multiply the total kernel response of leaf node p by the signal strength of leaf node p to obtain the weighted contribution value of leaf node p; finally, normalize the weighted contribution values ​​of all leaf nodes so that the sum of all leaf node weights is 1, thus obtaining the leaf node weight set. The introduction of signal strength is to highlight the contribution of strong signal nodes to the predicted coordinates. In the finger touch scenario, nodes with high signal strength usually correspond to the core area of ​​the contact area, and their positions are closer to the final stable coordinates. In the passive pen touch scenario, nodes with high signal strength are high gradient nodes, and focusing on these nodes can avoid coordinate shifts caused by edge effects.

[0057] The third step is to generate the predicted coordinates. The predicted coordinates refer to the estimated final stable contact position during the initial contact phase. Their calculation is based on the geometric boundary center coordinates of the leaf nodes and the leaf node weights. The specific calculation method is as follows: First, obtain the geometric boundary center coordinates of each leaf node. These coordinates are determined by calculating the range of row electrode numbers and column electrode numbers within the leaf node's geometric boundary. Specifically, the median of the row electrode number range is taken as the x-direction coordinate, and the median of the column electrode number range is taken as the y-direction coordinate. Second, multiply the x-direction coordinate of each leaf node by its weight to obtain the weighted x-direction coordinates of all leaf nodes. Summate these weighted x-direction coordinates to obtain the x-component of the predicted coordinates. Using the same method, multiply the y-direction coordinate of each leaf node by its weight and sum them to obtain the y-component of the predicted coordinates. Combining the x-component and y-component yields the predicted coordinates. These coordinates belong to a two-dimensional real number space and can directly correspond to the physical position of the touchscreen.

[0058] The fourth step is to modulate the confidence level. The confidence level is a parameter used to measure the reliability of the predicted coordinates, with a value ranging from [0,1]. The closer the value is to 1, the higher the consistency between the predicted coordinates and the final stable coordinates; the closer the value is to 0, the lower the reliability of the predicted coordinates. The confidence level modulation needs to be determined based on the product of the monotonically decreasing prediction variance and the prior factor, ensuring that the higher the prediction variance, the lower the confidence level, and the lower the prediction variance, the higher the confidence level.

[0059] Specifically, the method for calculating the prediction variance is as follows: First, based on the leaf node weight set and the coordinates of the geometric boundary center of the leaf nodes, calculate the dispersion of all leaf node coordinates relative to the predicted coordinates; second, multiply the squared Euclidean distance between the coordinates of each leaf node and the predicted coordinates by the weight of that leaf node, and sum the results to obtain the prediction variance. A larger prediction variance indicates a higher degree of dispersion of the leaf node coordinates, lower reliability of the predicted coordinates, and a correspondingly lower confidence level; a smaller prediction variance indicates a more concentrated distribution of the leaf node coordinates, higher reliability of the predicted coordinates, and a correspondingly higher confidence level.

[0060] Specifically, the method for determining the prior factor is as follows: the prior factor is a correction coefficient determined based on historical data of the touch scenario and the characteristics of the current scenario, with a value range of [0,1]. For low-temperature environment scenarios, the prior factor is determined through historical low-temperature drift statistics, that is, by statistically analyzing the deviation rate between the predicted coordinates and the actual stable coordinates in different low-temperature ranges. The higher the deviation rate, the smaller the value of the prior factor, in order to reduce the confidence of unreliable predictions in low-temperature environments. For passive pen or non-conductive touch tool scenarios, the prior factor is determined through historical low-coupling touch data, that is, by statistically analyzing the deviation of the predicted coordinates when using passive pen touch. The larger the deviation, the smaller the value of the prior factor. For normal finger touch scenarios, the prior factor takes a default value, which is determined through the prediction accuracy of a large number of normal touch samples to ensure that the confidence level can accurately reflect the prediction reliability under the condition of no special interference. Meanwhile, the determination of prior factors also needs to be combined with the contact morphology labels of the dynamic contact quadtree. If the contact morphology label is dot-shaped, it indicates that the current situation is passive pen touch control, and the prior factors of the low coupling scene need to be called. If the spatial distribution of leaf nodes shows edge concentration characteristics (judged by the directional distribution of spatial gradient vectors), it indicates that the current situation may be in a low temperature environment, and the prior factors of the low temperature scene need to be called.

[0061] Specifically, the monotonically decreasing mapping method based on prediction variance is as follows: An exponential decay mapping is used to convert the prediction variance into a basic reliability coefficient within the range of [0,1]. The calculation formula is: basic reliability coefficient = exp(-λ·prediction variance), where λ is the decay coefficient, used to adjust the influence of prediction variance on the basic reliability coefficient. The value of λ is determined through calibration using a large number of touch samples. Specifically, prediction variance and actual coordinate deviation data are collected under different scenarios. With the goal of "the basic reliability coefficient is 1 when the prediction variance is 0, and the basic reliability coefficient approaches 0 when the prediction variance reaches its typical maximum discrete value," the value of λ is iteratively optimized to ensure that the basic reliability coefficient accurately reflects the negative correlation between prediction variance and actual reliability.

[0062] The prediction variance is multiplied by the prior factor to obtain the confidence value. If the calculated confidence value is lower than 0, then 0 is taken as the final confidence value; if it is higher than 1, then 1 is taken as the final confidence value, ensuring that the confidence value meets the range requirement of [0,1].

[0063] Finally, the generated predicted coordinates are combined with the modulated confidence scores to form a stable coordinate prediction mapping output. The stable coordinate prediction mapping is based on the input of the current state of the dynamic contact quadtree, outputting the corresponding predicted coordinates and confidence scores. This mapping relationship is the same as the initial position prediction algorithm for contact dynamics sensing described above. This mapping relationship satisfies the local Lipshitz continuity condition, ensuring that small changes in the dynamic contact quadtree do not lead to drastic fluctuations in the prediction results, providing continuous and reliable input for the stable decision-making in step three.

[0064] For example, taking a passive pen touch scenario, the workflow of step two is as follows: In the dynamic contact quadtree constructed in step one, the contact morphology label of the leaf nodes is point-like, and the leaf nodes with high signal strength are concentrated in a small area (corresponding to the point-like contact of the passive pen); after mapping the dynamic contact quadtree to the local Lipshitz continuous mapping space, when performing adaptive kernel regression, the similarity function will give the high signal strength leaf nodes (i.e., high gradient nodes) and other nodes higher kernel response values, making the weight of these nodes significantly higher than that of edge nodes; when generating predicted coordinates, the high-weight high gradient nodes become dominant, avoiding coordinate offset caused by edge effects; when modulating the confidence, the prior factors of the passive pen scenario are called. If historical data shows that the prediction deviation is small in this type of scenario, the prior factor value is higher, and finally, the predicted coordinates with high confidence are output, forming a stable coordinate prediction mapping.

[0065] The role and effect of step two are mainly reflected in the following aspects: Traditional touch systems' centroid method only calculates coordinates based on the signal strength of the current frame, without considering the evolution of the signal from weak to strong and the contact from point to surface, causing the coordinates in the initial contact stage to jump with signal changes. This step maps the dynamic contact quadtree to the mapping space and constructs a similarity function by combining differential geometric features such as time derivative and spatial gradient vector. This allows the predicted coordinates to not only reflect the current signal distribution but also incorporate information about the signal evolution trajectory. For example, when a finger touches the screen, the algorithm predicts the core position after the contact area expands based on the changing trend of the time derivative, outputting stable coordinates in advance and avoiding the deviation between the first frame coordinates and subsequent stable coordinates, fundamentally alleviating the initial point jump phenomenon in the main problem. Adaptive kernel regression enhances scene adaptability. For loosely coupled touch tools such as passive pens, the algorithm prioritizes high-signal-strength nodes with stable temporal derivatives and concentrated spatial gradients by prioritizing node signal distance, assigning them higher weights to ensure that predicted coordinates focus on the contact core rather than the edge region, avoiding coordinate shifts caused by weak signals. Simultaneously, reliable coordinates can be output without extending signal accumulation time, reducing user pressing or pausing operations and restoring smooth operation. For low-temperature environments, the algorithm incorporates historical low-temperature drift data through prior factors, combined with node signal distance to suppress the weight of edge nodes (edge ​​node drift is more significant at low temperatures). Furthermore, it utilizes the continuity of the mapping space to ensure that predicted coordinates are not disturbed by the nonlinear rising edge of the signal, solving the problem that spatial inconsistency drift at low temperatures is difficult to compensate for with fixed calibration, ensuring the accuracy of first-point positioning in automotive and outdoor industrial control scenarios. The sparsity of the dynamic contact quadtree and the low complexity of the algorithm meet the resource constraints of embedded platforms. The dynamic contact quadtree constructed in step one contains only a small number of leaf nodes (usually less than 20), and the adaptive kernel regression process only needs to calculate these leaf nodes, without traversing all sensing units of the touch sensor. Meanwhile, all calculations can be performed using fixed-point integers, eliminating the need for floating-point units. This results in low execution time and memory usage on embedded platforms such as automotive MCUs and industrial HMIs, preventing increased touch latency due to excessive algorithm complexity and ensuring end-to-end response speed meets the real-time requirements of high-reliability scenarios. Dynamic modulation of confidence provides a quantitative basis for multimodal decision-making in step three, improving interaction reliability. Traditional systems only output coordinates without reliability assessment, leading to blind decision-making based on "reporting points whenever a signal is available," which can easily result in outputting incorrect coordinates before the signal stabilizes. This step combines prediction variance with prior factors to attach a confidence level to each predicted coordinate, allowing step three to flexibly select a decision strategy based on confidence. High confidence results in immediate coordinate output, avoiding delays; low confidence results in delayed confirmation or enhanced verification, preventing misjudgments. This quantitative basis ensures a more targeted decision-making process, reducing the probability of accidentally clicking adjacent buttons or misjudging clicks and swipes, further improving interaction security in automotive, medical, and other scenarios.

[0066] Step 3: Based on the stable coordinate prediction mapping, perform multimodal first point decision and touch event generation, output touch events that conform to human-computer interaction semantics and feed them back to the operating system;

[0067] First, a parsing operation of the stable coordinate prediction mapping is performed to extract the predicted coordinates and confidence scores. The output of the stable coordinate prediction mapping includes the predicted coordinates (denoted as...). ) and confidence (denoted as conf), where the predicted coordinates The confidence level (conf) ranges from [0,1] and is used to quantify the reliability of the predicted coordinates, corresponding to the position coordinates on the physical plane of the touchscreen. The parsing process requires the inverse operation of the mapping to transform the mapping relationship stored in the stable coordinate prediction mapping into specific numerical predicted coordinates and confidence levels, ensuring that the subsequent decision-making process can be directly executed based on the quantified parameters. After parsing, contact morphology labels need to be extracted from the dynamic contact quadtree constructed in step one. The contact morphology labels are parameters representing the contact type stored in the leaf nodes of the dynamic contact quadtree, with values ​​of point, area, or edge-dominated, used to determine the type of tool used in the current touch operation, such as a passive pen corresponding to point and a finger corresponding to area, providing scene recognition basis for multimodal decision-making.

[0068] Secondly, a preset threshold τ is determined as the benchmark parameter for confidence comparison. The preset threshold τ is a critical value used to divide the high and low confidence intervals, and its value needs to be determined through statistical calibration of a large number of touch scenario samples. Specifically, the method for determining the preset threshold τ is as follows: collect sample data from different touch scenarios (including operation of fingers, gloved fingers, passive pens, and non-conductive touch tools in normal and low temperature environments), calculate the deviation between the confidence of the predicted stable coordinate mapping and the actual stable coordinate for each sample; with the goal of "when the confidence ≥ τ, the deviation between the predicted coordinate and the actual stable coordinate is lower than the allowable error threshold of the touch system", the value of τ is iteratively adjusted to find the minimum τ value that can cover the vast majority of high-reliability samples, which is then used as the final preset threshold τ. This threshold needs to be stored in the touch system's configuration file, supporting fine-tuning according to the reliability requirements of different application scenarios (such as in-vehicle central control, industrial HMI), ensuring that the value of τ is higher in high-reliability scenarios to prioritize coordinate accuracy; and that the value of τ can be appropriately reduced in ordinary scenarios to balance response speed and accuracy.

[0069] After parameter preparation is completed, the multimodal first-point decision logic is initiated. The multimodal first-point decision is based on the comparison result of the confidence level conf and the preset threshold τ, the contact morphology label, and the system operating environment status, and the decision strategy is executed according to the scenario.

[0070] First, the high-confidence click event triggering strategy. When the confidence level conf ≥ the preset threshold τ, it indicates that the predicted coordinates output by the stable coordinate prediction mapping are... The reliability meets system requirements, and touch events can be triggered directly without additional verification. Specifically, the predicted coordinates... The coordinates are marked as final valid coordinates, and the touch event generation module is invoked to generate a high-confidence click event. This high-confidence click event represents a single reliable touch operation by the user, and its attributes include the final valid coordinates, event trigger time, and confidence level (marked as high confidence). It informs the operating system that the accuracy of the current touch operation has been confirmed, and the corresponding interactive command, such as opening an application or triggering a button function, can be executed directly. The decision rule for this strategy is: execution is only performed when conf ≥ τ and the touch shape label is not abnormal (e.g., there are no contradictory labels of both point and area shapes), to avoid false triggering due to label abnormalities.

[0071] Second, the delayed confirmation mode activation strategy. When the confidence level conf < the preset threshold τ but the signal continues to grow, it indicates that the reliability of the current predicted coordinates has not yet met the standard. However, the signal in the initial contact stage is still evolving, such as the continuous expansion of the finger contact area and the enhancement of coupling capacitance. Further verification of coordinate stability is required through a delay. Specifically, the method for judging the continuous signal growth is as follows: extract the time derivative (dΔC / dt) of the leaf nodes of the dynamic contact quadtree in step one. If the average time derivative of the current sampling period is greater than the signal growth threshold, it is determined that the signal is continuously growing. The signal growth threshold is the critical value that characterizes the signal in the rising stage, and is determined by the maximum value of the signal fluctuation in the non-contact state. After activating the delayed confirmation mode, the system needs to wait for the next sampling period and re-execute the construction of the dynamic contact quadtree in step one and the generation of stable coordinate prediction mapping in step two to obtain the predicted coordinates of the next frame. Calculate the predicted coordinates of the current frame using the confidence level 'conf'. With the predicted coordinates of the next frame The Euclidean distance is used to determine if the trajectory convergence is less than the trajectory convergence threshold. As the final valid coordinates, a delayed confirmation click event is generated. The trajectory convergence threshold is the critical value for judging whether the predicted coordinates are stable. It is determined by the 95th percentile (P95) of the coordinate fluctuation values ​​of adjacent frames in the stable phase. This ensures that when the fluctuation amplitude of the predicted coordinates in multiple consecutive frames is less than this threshold, it is judged as coordinate convergence. The 95th percentile (P95) means that 95% of the coordinate fluctuation values ​​of adjacent frames in the stable phase are less than this threshold, and only 5% of the extreme fluctuation values ​​exceed it, avoiding the threshold being too lenient due to a few maximum fluctuation values. If the distance is still greater than the trajectory convergence threshold, the system continues to wait for the next frame for verification until the coordinates converge in two consecutive frames or the number of waiting times reaches the maximum delay number. If convergence is not achieved after exceeding the maximum delay number, an invalid touch event is generated to avoid long periods of no response. The maximum delay number is determined according to the real-time requirements of the touch system to ensure that the total delay time does not exceed the upper limit of the delay that the user can perceive.

[0072] Third, the passive pen enhancement mode activation strategy. When the detected contact pattern label is dot-shaped and the confidence level (conf) is at a medium level, it indicates that the current touch tool is a passive pen or a non-conductive touch tool. Its contact pattern is fixed and the signal is weak, making it prone to prediction coordinate offset due to edge effects. Coordinate fusion is needed to improve accuracy. The medium level refers to conf falling within the range [τ1, τ), where τ1 is the lower limit threshold of medium confidence. The value of τ1 is determined through sample statistics of passive pen touch scenarios to ensure coverage of the typical confidence range during passive pen operation. Specifically, the execution steps of this strategy are as follows:

[0073] (1) Obtain the coordinates of the maximum response node. From the leaf nodes of the dynamic contact quadtree constructed in step one, select the leaf node with the largest signal strength, and define the coordinates of the geometric boundary center of the leaf node as the coordinates of the maximum response node (denoted as (x_max, y_max)). The leaf node with the largest signal strength corresponds to the actual contact core area between the passive pen and the touch screen. Its coordinates are least affected by the edge effect and can be used as the reference for coordinate fusion.

[0074] (2) Perform coordinate weighted fusion. Map the stable coordinate predictions to the predicted coordinates output. The coordinates of the node with the maximum response (x_max, y_max) are weighted and fused to obtain the final effective coordinates (x_fuse, y_fuse). Specifically, the weight coefficient 'a' in the weighted fusion is used... The values ​​of (x_max, y_max) and (b) are determined through training on samples from the passive pen scene. The value of b is greater than that of a to ensure the core role of the coordinates of the maximum response node and to suppress the offset caused by edge effects. The weight coefficients must satisfy a+b=1 to avoid the coordinates from exceeding the physical range of the touch screen after fusion.

[0075] (3) Generate passive pen enhanced click events. Pass the final effective coordinates (x_fuse, y_fuse) after fusion to the touch event generation module to generate passive pen enhanced click events. The attributes of the event are marked as passive pen, which makes it easier for the operating system to optimize the interaction logic for passive pen operations, such as adjusting the cursor size and enhancing the smoothness of the handwriting.

[0076] While implementing the above decision-making strategy, it is necessary to simultaneously determine whether the system is operating at low temperatures. If it is, non-uniform offset compensation is performed on the predicted coordinates to correct the spatial inconsistency drift caused by the low temperature. Specifically, the method for inferring the system's low-temperature operating state is as follows:

[0077] Method 1: Inference based on backlight power consumption. Real-time power consumption of the touchscreen backlight module is collected. Based on the current real-time brightness setting and driving mode of the touchscreen, the collected real-time power consumption is normalized and compared with the standard backlight power consumption under normal temperature conditions. If the normalized real-time power consumption is lower than the standard backlight power consumption and the difference is greater than the power consumption deviation threshold, it is inferred to be operating in a low-temperature state. The standard backlight power consumption is the average power consumption of the touchscreen under normal temperature and standard brightness settings, determined through factory calibration. The power consumption deviation threshold is a critical value characterizing the impact of low temperature on backlight power consumption, determined through power consumption testing under low-temperature conditions.

[0078] Method 2: Inference based on historical operating data. The mean signal intensity and spatial gradient vector distribution of the signals from the dynamically contacted leaf nodes of the quadrilateral tree in Step 1 are extracted and matched against a pre-stored low-temperature signal feature library. If the matching degree is higher than the feature matching threshold, it is inferred to be a low-temperature operating state. The low-temperature signal feature library stores signal intensity ranges and gradient distribution patterns under different low-temperature intervals, collected through touch experiments in a low-temperature environment. The feature matching threshold is a critical value characterizing the similarity between the current signal features and low-temperature features, determined through sample statistics.

[0079] When a low-temperature operating state is inferred, a pre-stored low-temperature calibration mapping table is invoked to compensate for the predicted coordinates. The low-temperature calibration mapping table is a compensation parameter table divided according to the touchscreen area. Its structure is as follows: the physical plane of the touchscreen is divided into several compensation areas according to a preset grid, such as a 10×10 grid. Each compensation area corresponds to a record, including the area number, the area center coordinates, and the x-direction and y-direction compensation amounts for that area in different low-temperature ranges. The pre-storage method for the low-temperature calibration mapping table is as follows: in a low-temperature environment laboratory, the ambient temperature is gradually reduced from room temperature to a target low-temperature range, such as -10°C to -30°C. At each temperature point, multiple touch operations are performed at the center position of each compensation area using a standard touch tool, such as a precision stylus. The deviation between the predicted coordinates of the stable coordinate prediction mapping and the actual coordinates of the standard touch tool is recorded. The average deviation is taken as the compensation amount for that area at that temperature point. The table is organized according to the temperature range and area number to form the low-temperature calibration mapping table, which is stored in the non-volatile memory of the touch system.

[0080] Specifically, the non-uniform offset compensation method is as follows: based on the current predicted coordinates... Alternatively, the fused coordinates (x_fuse, y_fuse) are used to determine the compensation region to which it belongs (by comparing the coordinates with the boundary range of each compensation region); the x-direction compensation amount Δx and y-direction compensation amount Δy of the compensation region under the current low temperature range are extracted from the low temperature calibration mapping table; the x-comp of the predicted coordinates is added to Δx, and the y-comp is added to Δy to obtain the final effective coordinates after compensation (x_comp, y_comp), ensuring that the coordinate drift under low temperature is corrected.

[0081] Finally, the touch event generation and feedback operations are performed. Based on the above decision-making strategy and compensation processing results, the touch event generation module needs to generate touch events that conform to human-computer interaction semantics. The event types include high-confidence click events, delayed confirmation click events, passive pen-enhanced click events, long press events, and invalid contact events. Among them, the triggering condition for the long press event is as follows: In delayed confirmation mode, if the predicted coordinates converge for multiple consecutive frames, the absolute value of the time derivative of the near-steady state signal is less than the signal steady-state threshold, and the duration of the touch operation exceeds the long press threshold, which is the critical time representing the user's long press intention and is determined according to industry-standard interaction, then a long press event is generated. The signal steady-state threshold is the critical value representing the signal stopping significant changes. It is determined by statistically analyzing the fluctuation range of the time derivative during the stable touch phase, ensuring that the signal is considered near steady state only when it no longer increases or decreases. The triggering conditions for invalid touch events are: the delayed confirmation mode fails to converge after exceeding the maximum delay count, the touch pattern labels are contradictory (e.g., point and area labels exist simultaneously), or the signal strength is lower than the touch trigger threshold (the touch trigger threshold is the minimum signal strength representing a valid touch, determined through contactless noise statistics). This is used to inform the operating system that the current operation cannot form a valid touch, preventing erroneous command execution. After a touch event is generated, the event data (including final valid coordinates, event type, confidence level, and trigger time) is fed back to the operating system via the system bus, completing the entire initial decision-making and event generation process.

[0082] For example, taking the scenario of passive pen touch control of the vehicle's central control screen in a low-temperature environment as an example, the workflow of step three is as follows: The system parses the stable coordinate prediction mapping to obtain the predicted coordinates. The system is compared with the confidence level conf (located in the [τ1,τ) range), and the contact morphology labels are extracted as point-like from the dynamic contact quadtree. The system is inferred to be operating at low temperature based on backlight power consumption and historical data. The passive pen enhancement mode is activated, and the coordinates of the maximum response node (x_max, y_max) are obtained. Weighted fusion is performed to obtain (x_fuse, y_fuse). The low-temperature calibration mapping table is queried based on (x_fuse, y_fuse) to obtain the corresponding compensation region's Δx and Δy, and the compensated (x_comp, y_comp) is calculated. Since conf is at a medium level and fusion and compensation have been completed, a passive pen enhancement click event is generated, and (x_comp, y_comp) and event information are fed back to the vehicle operating system to ensure the accuracy of the first point of passive pen touch control in vehicle scenarios.

[0083] The main functions and effects of this step are reflected in the following aspects: It solves the initial point jump problem in the main issue, meeting the requirements of high-reliability scenarios. The traditional system's serial logic of "outputting as soon as the signal meets the standard" leads to the direct output of unstable coordinates during the initial contact phase, resulting in an initial point jump. This step, through multimodal decision logic, uses stable coordinate prediction mapping as the decision center. With high confidence, it directly outputs reliable coordinates, avoiding delay; with low confidence, it delays confirmation until coordinate convergence, rejecting the output of unstable coordinates. Simultaneously, combined with contact pattern labeling and low-temperature compensation, it ensures that the initial point coordinates match the final stable position in different scenarios, eliminating the root cause of the initial point jump from the decision-making level. This enables "one-touch accuracy" in scenarios such as vehicle central control and industrial HMIs, avoiding accidental clicks on adjacent buttons or triggering incorrect commands. For the weak signals and point-like contact characteristics of passive pens and non-conductive touch tools, traditional solutions either have no response or output offset coordinates, forcing users to extend operation time. This step identifies point-like contacts through contact pattern tags and enables a passive pen enhancement mode: it fuses predicted coordinates based on the coordinates of the maximum response node (contact core) to suppress offsets caused by edge effects; simultaneously, it can improve confidence through fusion without waiting for signal accumulation to reach high intensity, shortening response time. Users do not need to press hard or prolong the pause, restoring the smoothness of operation to a level close to finger touch, filling the gap in existing solutions for optimizing the first-point response of loosely coupled touch tools. Coordinate drift in low-temperature environments cannot be solved by traditional fixed offset calibration due to spatial non-uniformity. This step infers the low-temperature operating state from multiple dimensions and performs regional compensation in conjunction with a pre-stored low-temperature calibration mapping table. The compensation amount for each screen area is determined based on actual low-temperature experimental data, accurately matching the drift characteristics of different areas; at the same time, the compensation process is embedded in the decision-making process, requiring no additional hardware modules, ensuring that the first-point positioning accuracy at low temperatures is consistent with that at room temperature, solving the problem of performance degradation of equipment in winter in low-temperature scenarios such as automotive, outdoor, and industrial applications, and expanding the environmental adaptability of the touch system. The multimodal decision-making in this step is implemented based on a purely rule-driven state machine, without AI inference or iterative computation. All judgments, such as the comparison of conf and τ and signal growth determination, are simple numerical calculations. Low-temperature compensation only requires table lookup and addition operations, and the execution time on a typical automotive MCU is far below the real-time threshold. Meanwhile, the decision-making strategy, through the design of "high-confidence priority output and low-confidence delay verification," ensures reliability while avoiding excessive latency. End-to-end touch latency is controlled within the user's perceptible range, fully meeting the requirements of embedded high-real-time scenarios. Multi-scenario adaptive output achieves "one model covering all scenarios," reducing system complexity. Traditional solutions require separate compensation modules for passive pens and low-temperature environments, resulting in complex system structures and poor compatibility.This step integrates the decision-making logic of all scenarios into a single multimodal framework. By using three dimensions—confidence of stable coordinate prediction mapping, contact form label, and environmental state—it automatically adapts to all scenarios, including fingers, passive pens, low temperatures, and normal temperatures, without requiring additional module switching. At the same time, the event attributes carry scenario identifiers, such as high confidence for passive pens, which facilitates targeted adaptation by the operating system. This achieves closed-loop control of "one model, all scenarios, and self-adaptation," reducing system development and maintenance costs and improving the versatility of the touch system.

[0084] Example 2:

[0085] This embodiment, based on Embodiment 1, provides a capacitive touch signal adaptive compensation device, such as... Figure 4 As shown, the device is characterized in that it stores a computer program adapted for loading by a processor to execute a capacitive touch signal adaptive compensation method as described above, and the device comprises:

[0086] Data acquisition and quadtree construction unit: used to acquire raw capacitance signal grid data during the initial contact phase and construct a dynamic contact quadtree;

[0087] Mapping construction unit: used to run the contact dynamics-aware first point position prediction algorithm based on the dynamic contact quadtree to construct a stable coordinate prediction mapping with confidence.

[0088] Decision and event generation unit: Based on the stable coordinate prediction mapping, it performs multimodal first point decision and touch event generation, outputs touch events that conform to human-computer interaction semantics and feeds them back to the operating system.

Claims

1. A capacitive touch signal adaptive compensation method, characterized in that, The method includes: Step 1: Collect raw capacitance signal grid data during the initial contact phase and construct a dynamic contact quadtree; Step 2: Based on the dynamic contact quadtree, run the contact dynamics-aware first point position prediction algorithm to construct a stable coordinate prediction mapping with confidence. Step 3: Based on the stable coordinate prediction mapping, perform multimodal first point decision and touch event generation, output touch events that conform to human-computer interaction semantics and feed them back to the operating system; The structural features of the dynamic contact quadtree include: a root node, child nodes, and leaf nodes; the root node corresponds to the initially determined single non-zero response region; the child nodes correspond to the sub-regions generated during the recursive quadtree partitioning process; and the leaf nodes correspond to the smallest sub-regions formed after the recursive quadtree partitioning terminates. The steps of the initial position prediction algorithm for contact dynamics sensing are as follows: Construct a similarity function and calculate the kernel response value based on the lowest common ancestor depth between leaf nodes and the node signal distance; The weight of each leaf node is calculated and normalized by combining the sum of the kernel response values ​​of each leaf node with the signal strength; the predicted coordinates are obtained by summing the geometric boundary center coordinates of the leaf node with the corresponding weights.

2. The adaptive compensation method for capacitive touch signals according to claim 1, characterized in that, The method for constructing the dynamic contact quadtree is as follows: The non-zero response region is determined based on a two-dimensional matrix of initial contact capacitance change. The non-zero response region refers to the continuous area on the touch screen formed by the sensing units corresponding to the elements in the two-dimensional matrix of initial contact capacitance change whose element values ​​are greater than the static noise threshold. For each non-zero response region, a recursive quadtree partition is performed, and each node of the dynamic contact quadtree is assigned geometric boundary, signal strength, time derivative, spatial gradient vector, and contact morphology label attributes.

3. The adaptive compensation method for capacitive touch signals according to claim 2, characterized in that, The two-dimensional matrix of initial contact capacitance change is as follows: the row dimension corresponds to the row electrode number of the touch sensor, and the column dimension corresponds to the column electrode number of the touch sensor; each element in the two-dimensional matrix of initial contact capacitance change is the average capacitance change of the corresponding sensing unit in the first N sampling periods.

4. The adaptive compensation method for capacitive touch signals according to claim 3, characterized in that, The recursive quadrilateral partitioning method is as follows: Set the non-zero response region as the current region to be subdivided, and determine the geometric boundary of the region; Calculate the signal variance of the current region to be partitioned; if the signal variance exceeds the dynamic noise threshold, partition the region into four equal sub-regions; if the signal variance is less than or equal to the dynamic noise threshold, stop partitioning the region and use the region as a leaf node of the dynamic contact quadtree. Repeat the above operation for each sub-region until the signal variance of all sub-regions is less than or equal to the dynamic noise threshold, then stop the recursive subdivision.

5. The adaptive compensation method for capacitive touch signals according to claim 4, characterized in that, The method for calculating the node signal distance is as follows: select any two leaf nodes from the dynamic contact quadtree as the calculation objects; obtain the geometric boundary center coordinates of the two leaf nodes and calculate the Euclidean distance between the coordinates; obtain the signal strength of the two leaf nodes and calculate the absolute difference of the signal strength; perform weighted correction on the Euclidean distance and the absolute difference of the signal strength, and introduce a distance correction coefficient determined based on the difference in signal variance of the non-zero response regions to which the two nodes belong, and finally obtain the node signal distance.

6. The adaptive compensation method for capacitive touch signals according to claim 5, characterized in that, The method for constructing the stable coordinate prediction mapping is as follows: The leaf node parameters of the dynamically contacting quadtree are used as input, including the geometric boundary center coordinates, signal strength, time derivative, and spatial gradient vector. Substitute the leaf node parameters into the initial position prediction algorithm of contact dynamics sensing to calculate the predicted coordinates. The confidence score is obtained based on the prediction variance and prior factors; The stable coordinate prediction mapping is based on the input of the leaf node parameters of the dynamic contact quadtree, and the output of the corresponding predicted coordinates and confidence levels.

7. The adaptive compensation method for capacitive touch signals according to claim 6, characterized in that, The method for multimodal first point decision is as follows: if the confidence level of the stable coordinate prediction mapping is not lower than a preset threshold, a high-confidence click event is directly triggered; if the confidence level is lower than the preset threshold but the original capacitance signal continues to increase, a delayed confirmation mode is started and the signal change is continuously monitored. If the contact pattern label is dot-shaped and the confidence level is medium, enable the passive pen enhancement mode to improve positioning accuracy.

8. A capacitive touch signal adaptive compensation device, characterized in that, The device stores a computer program adapted for loading by a processor to execute a capacitive touch signal adaptive compensation method as described in any one of claims 1-7, the device comprising: Data acquisition and quadtree construction unit: used to acquire raw capacitance signal grid data during the initial contact phase and construct a dynamic contact quadtree; Mapping construction unit: used to run the contact dynamics-aware first point position prediction algorithm based on the dynamic contact quadtree to construct a stable coordinate prediction mapping with confidence. Decision and event generation unit: Based on the stable coordinate prediction mapping, it performs multimodal first point decision and touch event generation, outputs touch events that conform to human-computer interaction semantics and feeds them back to the operating system.

Images