An active infrared interactive target tracking method based on edge computing

Abstract

The application discloses an active infrared interactive target tracking method based on edge calculation, and particularly relates to the technical field of infrared interactive target tracking, and is used for solving the problem that the existing active infrared interactive target is easily confused with a projection change background under the conditions of dynamic projection pictures, local brightness change and on-site environmental disturbance, leading to inaccurate identification, unstable tracking and mapping jump. The projection change reference field is generated by collecting the infrared reference frame and the projection change reference frame of the target interactive surface, the infrared difference enhancement is performed on the current infrared image frame according to the local response unit, the frame displacement correction under the continuous observation window, the peak value stability correction and the background inversion correction are combined, the active infrared interactive target is continuously confirmed, the real interactive target is effectively distinguished from the projection change background under the coexisting conditions of the projection picture change, the local brightness fluctuation and the on-site interference, the misidentification and the missed identification are reduced, and the tracking stability is improved.

Description

Technical Field

[0001] This invention relates to the field of infrared interactive target tracking technology, specifically to an active infrared interactive target tracking method based on edge computing. Background Technology

[0002] In existing immersive projection booths, interactive facade screens, and multi-material non-touch interactive interfaces, infrared image acquisition is commonly used to obtain the position of interactive elements on the target interactive surface. The recognition results are then converted into corresponding interactive coordinates to drive light point feedback, content switching, or trajectory linkage. Related solutions typically use high-response areas in infrared images as the basis for interactive target recognition, combining fixed threshold segmentation, background subtraction, region connectivity extraction, or adjacent frame position comparison to determine the presence of an interactive target and further complete coordinate mapping output. Under conditions of static backgrounds, single-material surfaces, or minimal changes in projected content, this method can achieve basic interactive detection and position output, and therefore has been applied in interactive displays, educational demonstrations, and multimedia control scenarios.

[0003] However, in actual exhibit operation, the target interactive surface often simultaneously carries dynamic projected images, and the image content exhibits local brightness abrupt changes, overall texture changes, and continuous motion transitions. Furthermore, the target interactive surface itself may have different surface features such as coating, frosted, or curved surfaces, and ambient lighting, local reflections, and short-term occlusions can also affect the infrared observation results. Under these conditions, existing solutions that rely solely on single-frame infrared high-response areas, fixed background apertures, or simple inter-frame differences for identification are prone to confusing local response fluctuations caused by projection changes with the actual interactive target, or causing candidate region drift, center position shift, and coordinate output jumps when the interactive target passes through bright edges, texture transition areas, and areas of local fluctuation. Especially during continuous movement of the interactive target, if the background aperture cannot maintain stable constraints with changes in the operating scene, subsequent target confirmation and coordinate output are easily affected by continuous interference from the dynamic projected content, thus impacting the continuity and stability of the interactive feedback.

[0004] Therefore, existing technologies still need to address a more specific problem: how to stably distinguish between real active infrared interactive targets and the changing projected background in operating scenarios where the dynamic projected image is constantly changing, local brightness fluctuates, and the on-site environment is disturbed, and how to maintain the continuity and stability of the interactive target tracking results and coordinate output results on this basis. This problem does not lie in simply improving infrared detection sensitivity, nor in optimizing coordinate mapping accuracy alone, but in how to balance the connection between background reference establishment, candidate extraction, continuous confirmation, center correction, mapping constraints, and output updates within the same operational chain, thereby reducing the impact of dynamic projected content on interactive target recognition and subsequent coordinate output. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an active infrared interactive target tracking method based on edge computing to solve the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an active infrared interactive target tracking method based on edge computing, comprising: S1. Acquire infrared reference frames and projection change reference frames of the target interaction surface, divide the local response units, and generate the projection change reference field. S2. Acquire the current infrared image frame, perform infrared difference enhancement according to the local response unit, and form an active infrared interactive candidate record; S3. Perform inter-frame shift verification, peak stability verification, and background inversion verification on active infrared interaction candidate records according to continuous observation windows to confirm valid interaction targets; S4. Extract the core response area and transition response area of ​​the effective interaction target, and generate the centroid correction value; S5. Generate target mapping coordinates based on the centroid correction amount, projection change reference field, and preset mapping matrix; S6. Perform rhythmic constraint output on the target mapping coordinates to generate structured tracking results, and update the projected change reference field according to local perturbation changes.

[0007] Furthermore, S1 includes: Infrared reference frames and projection change reference frames are acquired under the same target interaction surface, the same lens mounting position, the same exposure aperture, the same gain aperture, and the same rule version. Then, local response units are divided based on the effective imaging boundary of the target interaction surface; Then, for each local response unit, the infrared reference frame observation results are collected to form the basic infrared response lower limit and basic infrared response upper limit, and the projection change reference frame observation results are collected to form the response fluctuation lower limit and response fluctuation upper limit, so as to generate the projection change reference field.

[0008] Furthermore, S2 includes: Under the same lens mounting position, the same target interaction surface, and the same rule version, continuously acquire the current infrared image frame; Based on the local response unit, the observation part outside the invalid imaging boundary is excluded within the corresponding pixel range, and the remaining current infrared observation value is compared with the basic infrared response range determined by the basic infrared response lower limit and the basic infrared response upper limit, and the response fluctuation range determined by the response fluctuation lower limit and the response fluctuation upper limit in turn, and the infrared change part is retained as a candidate formation object. Candidate objects are then merged within the same local response unit to form a continuous coverage area, and the continuous coverage area is verified according to the closed candidate formation caliber to form an active infrared interactive candidate record.

[0009] Furthermore, S3 includes: Active infrared interactive candidate records are retrieved continuously according to the acquisition sequence number, and a continuous observation window is formed with the acquisition sequence number of the current active infrared interactive candidate record as the window endpoint. Within a continuous observation window, first perform inter-frame shift verification, then perform peak stability verification, and retain the candidate records corresponding to the end point of the continuous observation window that pass the peak stability verification to enter the background inversion verification.

[0010] Furthermore, S3 also includes: Perform background inversion verification on the candidate records corresponding to the endpoints of the continuous observation windows that have passed the peak stability verification; When the candidate record corresponding to the end point of the continuous observation window falls into the boundary position change range of the response fluctuation interval and cannot form a continuous boundary continuation chain, it is marked as a background reversal record. Candidate records corresponding to the endpoints of consecutive observation windows that are not marked as background inversion records are used as valid interaction target confirmation objects.

[0011] Furthermore, S4 includes: Under the same target interaction surface, the same lens mounting position, and the same rule version, retrieve the current infrared image frame, the effective target boundary position, the effective target center position, and the corresponding local response unit number corresponding to the effective interaction target; Then, within the pixel range covered by the corresponding local response unit number, extract the high-response continuous internal region that is continuously connected to the effective target center position as the core response region, and extract the continuous peripheral region located outside the core response region that is continuously connected to the core response region and lower than the corresponding region of the core response region as the transition response region. The main positioning center is then determined by the center of the range enclosed by the core response area boundary, and the centroid correction amount is generated based on the main distribution direction of the transition response area relative to the main positioning center.

[0012] Furthermore, S5 includes: Based on the effective target center position, apply the centroid correction amplitude according to the centroid correction direction to obtain the corrected positioning position; Next, determine the corresponding local response unit to which the corrected positioning position belongs, and perform pre-mapping constraint verification on the corrected positioning position based on the stable response range boundary of the corresponding local response unit determined by the projection change reference field.

[0013] Furthermore, S5 also includes: Using the corrected positioning position verified by pre-mapping constraints as the unique mapping reference, the preset mapping matrix record matching the target interaction surface number is called to determine the corresponding position of the corrected positioning position in the output coordinate plane and generate the target mapping coordinates.

[0014] Furthermore, S6 includes: The target mapping coordinates corresponding to the current acquisition sequence number are used as the current output candidate, and the current output position that was not transferred to the pending verification state before the current acquisition sequence number is retrieved as the previous valid output position. When the current position changes beyond the output rhythm boundary, the current output position is restricted to the allowable output range with the previous valid output position as the reference and the output rhythm boundary as the outer boundary. The structured tracking results are then regenerated, and the projected reference field is updated according to the local perturbation changes.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By acquiring infrared reference frames and projection change reference frames of the target interaction surface and generating a projection change reference field, infrared difference enhancement is performed on the current infrared image frame according to the local response unit. Combined with inter-frame displacement verification, peak stability verification and background reversal verification under continuous observation window, the active infrared interactive target is continuously confirmed. Thus, even under the conditions of continuous changes in the projected image, local brightness fluctuations and on-site environmental interference, the real active infrared interactive target can still be effectively distinguished from the projected changing background, reducing false recognition and missed recognition, and solving the technical problem in the prior art that interactive targets are easily interfered with by the dynamic content of the projection, resulting in unstable tracking results.

[0016] 2. By extracting the core response area and transition response area of ​​the effective interactive target, a centroid correction value is generated. Combined with the projection change reference field and the preset mapping matrix, the target mapping coordinates are generated. Then, rhythmic constraints are applied to the target mapping coordinates and the projection change reference field is updated according to the local disturbance changes. This improves the continuity, smoothness and scene adaptability of the interactive landing point mapping, so that the output results remain stable and traceable under the conditions of target movement, edge disturbance and local environmental changes. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of an active infrared interactive target tracking method based on edge computing; Figure 2 This is a schematic diagram of the process for generating a reference field for projection changes. Figure 3 This is a schematic diagram of the effective interactive target confirmation process under continuous observation window. Detailed Implementation

[0018] 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.

[0019] Example: Combined with Appendix Figures 1-3 This embodiment provides an active infrared interactive target tracking method based on edge computing, including: S1. Acquire infrared reference frames and projection change reference frames of the target interaction surface, divide the local response units, and generate a projection change reference field. The specific implementation is as follows: After the edge computing node is powered on and enters the calibration preparation period, the infrared image sensor installed at the corresponding observation position on the target interaction surface continuously acquires infrared reference frames. The target interaction surface is defined as the continuous imaging area that is completely covered by the field of view of the current infrared image sensor in a single deployment and serves as the object of single interaction coordinate mapping. It does not include decorative obstruction areas, areas outside splicing seams, areas obstructed by permanent reflective parts, and areas outside the effective imaging boundary. The execution is carried out by on-site debugging personnel in cooperation with the edge computing node. The acquisition period is limited to the period when the active infrared interaction target transmitter is turned off, no one enters the target interaction surface with an interactive component, the projection equipment maintains the benchmark standby screen or the predetermined benchmark playback segment according to the current deployment plan, and the on-site lighting maintains the standard lighting level after deployment confirmation.

[0020] Infrared reference frames are used to solidify the basic infrared response aperture of the target interaction surface under the current deployment conditions. Each frame contains the acquisition time, sensor identifier, lens mounting position, exposure aperture, gain aperture, target interaction surface number, acquisition batch number, and rule version identifier. The acquisition time is uniformly recorded using the local clock of the edge computing node. The exposure aperture, gain aperture, and rule version identifier are locked within this step and cannot be changed automatically. The rule version is the version identifier that locks the boundary conditions, update aperture, mapping aperture, and record field aperture involved in the current step. Under the same rule version, the judgment boundaries and record formation aperture remain unchanged. The source identifier is the unique identifier of the previous layer record that the current record directly inherits from, calls, or corresponds to, and is used to maintain the traceability of the cross-step record link.

[0021] After completing the infrared reference frame acquisition, projection change reference frames are acquired under the same target interaction surface, the same lens mounting position, the same exposure aperture, the same gain aperture, and the same rule version. The projection change reference frames are derived from the types of projection images that will appear when the current exhibit is officially running, and at least cover static low change images, local brightness change images, overall texture change images, and continuous motion images. If a certain type of image does not exist on site, it is marked in the record that the type does not exist and is not replaced by other images. The projection segment number, playback start and end time, image switching time, and on-site lighting status are written in simultaneously.

[0022] After the two types of reference frames are acquired, the records are merged according to a unified sorting order: first, they are sorted by acquisition time and time alignment is completed, with the local clock of the edge computing node as the reference; then, distorted frames are removed. Distorted frames are limited to frames that show continuous occlusion of the effective imaging area, regional saturation and brightness, abnormal boundary position shift due to lens shake, or failure to maintain the aperture of the current lens mounting position. Continuous occlusion refers to the situation where the same occluded area within the target interaction surface continuously occupies a fixed position within the effective imaging boundary and continuously covers the adjacent acquisition frames at that position. Regional saturation and brightness refers to the abnormal brightness state of a local area exceeding the current deployment confirmed brightness aperture within consecutive acquisition frames. Abnormal boundary position shift refers to the situation where the effective imaging boundary of the same target interaction surface cannot maintain the fixed boundary position after deployment confirmation in adjacent acquisition frames. The above-mentioned effective imaging boundary after deployment confirmation is based on the on-site calibration record after this installation is completed, and is not adjusted due to temporary changes in the image, short-term light fluctuations, or temporary observation habits of the debugging personnel.

[0023] Next, check the continuity of the acquisition sequence number. Only the record sequence number and missing markers are allowed to be filled in. Image content, lens mounting position, exposure aperture, gain aperture, and rule version identifier are not allowed to be filled in. If the continuous missing number exceeds the preset boundary, or if the lens mounting position, exposure aperture, gain aperture, or rule version identifier is missing, the processing will stop and the record will be transferred to the record to be checked. The record to be checked must include at least the target interaction surface number, acquisition batch number, missing field name, stop time, and stop reason.

[0024] After obtaining a valid reference frame, local response units are divided according to the effective imaging boundary of the target interaction surface. Each local response unit is limited to the smallest spatial partition obtained by dividing the effective imaging boundary of the target interaction surface according to a preset grid rule. Each local response unit has a fixed boundary, a unique number, does not overlap with each other, and continuously covers the entire effective imaging area without crossing invalid imaging areas, decorative obstruction areas, or permanent reflective obstruction areas. During the division, the effective imaging boundary of the target interaction surface is first confirmed. The effective imaging boundary adopts the boundary record jointly confirmed by the debugging personnel in this deployment based on the current field of view coverage, the actual observable range of the target interaction surface, and the on-site calibration record. After confirmation, the boundary is continuously divided into blocks to form a local response unit record with local response unit number, boundary position, corresponding pixel range, corresponding physical area identifier, acquisition batch number, and rule version identifier. During this deployment, the local response unit boundary remains unchanged. When the lens mounting position changes, the entire batch is rebuilt without using the old unit record. The old version record must not be directly overwritten by the new batch. Instead, the version relationship, formation time, device identifier, execution subject, and reconstruction reason are retained by appending the new version.

[0025] Based on local response units, infrared reference frames and projection change reference frames are merged one by one according to local response units to generate a projection change reference field. The projection change reference field is limited to a field reference record that records the basic infrared response range and projection change response fluctuation range of each local response unit under the current target interaction surface, current lens mounting position, and current rule version. The formation order is as follows: first, for each local response unit, the observation results of effective infrared reference frames are collected, and the corresponding distorted frame observation values ​​are removed according to a uniform filtering caliber to form the basic infrared response lower limit and basic infrared response upper limit of the local response unit. Then, the observation results of effective projection change reference frames are collected, and the frames are merged one by one according to the same local response unit boundary to form the response fluctuation lower limit and response fluctuation upper limit of the local response unit during the projection change. The limits, including the lower and upper limits, are formed by the corresponding set of observations after removing distorted frame observations, under the same rule version and according to a pre-locked boundary determination method. This is used to fix the stable response range of the local response unit under the current deployment conditions and serves as the basis for determining the boundary of the subsequent stable response range. It is not directly rewritten with instantaneous abnormal changes in a single frame. Finally, along with the target interaction surface number, local response unit number, lens mounting position identifier, acquisition batch number, rule version identifier, and formation time, it is written into the local storage area of ​​the edge computing node and submitted to the upper management terminal. The target interaction surface number, local response unit version number, projection change reference field version number, and formation completion status are submitted. The upper management terminal only writes back the reception status and version confirmation status. When the submission is interrupted, the local record is kept intact and the status is changed to pending submission.

[0026] This step is applicable to immersive projection imaging surfaces, multi-material non-touch interactive surfaces, and concealed installation observation scenarios. It is not applicable to situations where the target interactive surface is continuously displaced, the lens mounting position is not fixed, or the active infrared interactive target has entered the acquisition area.

[0027] Preferably, in the implementation of the exhibition booth with a circular imaging area of ​​1.8 meters, the infrared reference frame acquisition quantity can be set to 120 to 300 frames, the projection change reference frame covers 4 to 6 types of commonly used image segments, and 60 to 180 frames are acquired for each type. The local response unit is divided into 144 to 256 blocks. 30 local response units are randomly selected for inspection. It is required that the boundary deviation formed by the same batch does not exceed 1 pixel, and the version lock record completeness rate reaches 100%.

[0028] Without changing the technical concept of "acquiring infrared reference frames and projection change reference frames of the target interaction surface, dividing local response units, and generating projection change reference fields", the infrared image sensor can be replaced with a similar sensor that can output infrared image frames of the same spatial area under the same installation position and the same acquisition aperture. The target interaction surface can be replaced with a curved projection screen, a frosted plate surface, or a coated imaging surface.

[0029] S2. Acquire the current infrared image frame, perform infrared difference enhancement according to the local response unit, and form an active infrared interaction candidate record. The specific implementation is as follows: After generating the projection change reference field in the previous step, the edge computing node enters the data acquisition period. The infrared image sensor installed at the corresponding observation position on the target interaction surface continuously acquires the current infrared image frame. The current infrared image frame is limited to the real-time infrared observation frame continuously acquired by the edge computing node within the data acquisition period according to the locked lens mounting position, exposure aperture, gain aperture, and rule version, and used for infrared difference enhancement in this step. It does not take the formation of candidates as a prerequisite for naming. The acquisition position is still the target interaction surface confirmed in this deployment. The execution is automatically completed by the edge computing node. On-site debugging personnel only confirm the acquisition status after the initial connection, shutdown recovery, or sensor replacement. Each current infrared image frame is written with the acquisition time, sensor identifier, lens mounting position, target interaction surface number, acquisition sequence number, exposure aperture, gain aperture, and rule version identifier. The acquisition time is uniformly recorded by the local clock of the edge computing node. The exposure aperture, gain aperture, and rule version identifier are locked by the previous step and are not temporarily changed due to the current interaction action.

[0030] After acquiring the current infrared image frame, the frame is first organized according to the acquisition sequence number and acquisition time, with the acquisition sequence number as the primary factor and the acquisition time as the secondary factor to maintain the consistency of the frame order. This ensures that the current infrared image frame, the local response unit record, and the projection change reference field record are under the same lens mounting position, the same target interaction surface number, and the same rule version. For the organized current infrared image frame, only missing markers and continuity markers are allowed to be added; image content, lens mounting position, exposure aperture, gain aperture, and rule version identifier are not added. If the acquisition sequence number is continuously missing beyond the preset boundary, the lens mounting position is inconsistent, the rule version is inconsistent, the target interaction surface number is inconsistent, or the duplicate sequence number cannot correspond to a unique acquisition time, then the current round of candidate record formation is stopped, and the batch of frames is transferred to the pending verification record. This batch of frames will not participate in the current round of active infrared interaction candidate record formation.

[0031] Subsequently, infrared differential enhancement is performed according to the local response units. The local response units are recorded using the local response units with fixed boundaries and unique numbers from the previous step. Within each local response unit, the current infrared observation value is the infrared response record value of the current infrared image frame within the corresponding pixel range of that local response unit. Infrared differential enhancement is performed in a fixed order: first, the observation part outside the invalid imaging boundary is excluded within the corresponding pixel range of the local response unit; then, the remaining current infrared observation value is compared one by one with the basic infrared response interval corresponding to the same local response unit in the projection change reference field, and the stable background part falling into the basic infrared response interval is removed; then, the remaining current infrared observation value is compared one by one with the response fluctuation interval corresponding to the same local response unit, and the projection change part falling into the response fluctuation interval is removed. The infrared change part retained after the above two comparisons is used as the candidate formation object; here, the removal is limited to marking as a non-candidate source and not participating in the candidate merging in this step, without rewriting the original current infrared image frame record.

[0032] After infrared differential enhancement is completed, within the same local response unit, spatially adjacent candidate formation objects are merged into a continuous coverage area, while separate and unconnected candidate formation objects are merged into different continuous coverage areas without crossing local response units. The active infrared interaction target is the infrared action point emitted by the interaction device in the operation site. Its on-site manifestation on the target interaction surface is a local infrared high response area that moves with the operation. Continuous light spots and intermittent flashes that still fall within the acquisition aperture of the same rule version can all enter the candidate formation range of this step. The preset candidate boundary is the allowed boundary range of the candidate area locked according to the current rule version within a single local response unit. The diffusion area whose coverage exceeds the preset candidate boundary of a single local response unit, the drift area that cannot form a continuous coverage area boundary within the current infrared image frame, and the area that has fallen into the response fluctuation range of the projection change reference field are not considered as active infrared interaction target candidate sources.

[0033] Subsequently, each continuous coverage area is verified according to the closed candidate formation caliber. An active infrared interaction candidate record is formed when the following conditions are met: the continuous coverage area is within the effective imaging boundary of the target interaction surface; the continuous coverage area originates solely from measured observations of the current infrared image frame, not from completion markers; the continuous coverage area does not cross multiple non-adjacent local response units; and the continuous coverage area can form a unique candidate region boundary that is distinguishable from other continuous coverage areas within the same current infrared image frame. An active infrared interaction candidate record is formed when all the aforementioned conditions are met.

[0034] Active infrared interaction candidate records are limited to recording objects that are continuously covered areas within the current infrared image frame that may correspond to active infrared interaction targets. When they are formed, each item is written with the target interaction surface number, acquisition time, acquisition sequence number, local response unit number, candidate region boundary position, corresponding pixel range of the candidate region, candidate region center position, candidate region coverage area, current rule version identifier, and source frame identifier. The candidate region center position is uniformly determined under the same rule version according to the center position of the range surrounded by the candidate region boundary. The candidate region coverage area is recorded according to the pixel area of ​​the corresponding pixel range of the candidate region and stored in the local candidate record area of ​​the edge computing node for direct retrieval in the next step. Within the same current infrared image frame, candidate regions with overlapping or interconnected boundaries in the same local response unit form only one active infrared interaction candidate record and are marked as duplicated. Separate candidate regions form active infrared interaction candidate records separately. Candidates across frames are not merged in advance in this step.

[0035] This step is applicable to scenarios where active infrared interactive targets are extracted in real time during the operation of immersive projection platforms, interactive screens, and multi-material non-touch interactive surfaces. It is not applicable to situations where the lens mounting position has changed and recalibration has not been completed, the reference field has failed due to projection changes, or the target interactive surface has been continuously obscured.

[0036] Preferably, in the implementation of the exhibition booth with a circular imaging area of ​​1.8 meters, the current infrared image frame can be continuously acquired at 30 to 90 frames per second, and the number of local response units is maintained at 144 to 256. When 2 to 5 continuous coverage areas are formed in a single frame, the candidate record can be completed within 20 to 50 milliseconds. A random check of 300 consecutive frames is required, and the consistency rate between the active infrared interaction candidate record and the local response unit where the actual interaction point is located is not less than 95%.

[0037] Without changing the technical concept of "acquiring the current infrared image frame, performing infrared difference enhancement according to the local response unit, and forming an active infrared interactive candidate record", the infrared image sensor can be replaced with a similar sensor that outputs infrared image frames of the same spatial area under the same installation position and the same acquisition aperture.

[0038] S3. Perform inter-frame shift verification, peak stability verification, and background inversion verification on the active infrared interaction candidate records according to a continuous observation window to confirm the valid interaction targets. The specific implementation is as follows: After the edge computing node generates active infrared interaction candidate records in the previous step, it immediately retrieves active infrared interaction candidate records under the same target interaction surface, the same lens mounting position, and the same rule version according to the acquisition sequence number. It then uses a continuous observation window to sequentially verify this batch of active infrared interaction candidate records. The continuous observation window is limited to a fixed-length candidate comparison interval formed by continuously selecting a preset number of active infrared interaction candidate records forward from the acquisition sequence number of the current candidate record to be confirmed. If the fixed length is not reached, valid interaction target confirmation is not performed; only a pending observation mark is retained. The execution is automatically completed by the edge computing node, and the effective period is during the period when the active infrared interaction target continuously interacts with the target interaction surface. The source of the retrieved records is the local candidate record. The recording area must retrieve at least the target interaction surface number, acquisition time, acquisition sequence number, local response unit number, candidate region boundary position, candidate region center position, candidate region coverage area, source frame identifier, and duplicate merging mark. During retrieval, the data is first sorted by acquisition sequence number, and then the consistency of the target interaction surface number, lens mounting position associated record, and rule version identifier is checked. Only missing markers are allowed to be filled in; candidate region boundary position, candidate region center position, candidate region coverage area, and rule version identifier are not allowed to be filled in. If any record in the continuous observation window has an acquisition sequence number that exceeds the preset boundary, an inconsistent target interaction surface number, an inconsistent rule version, or a missing candidate region center position, the current continuous observation window verification is stopped, and the continuous observation window is transferred to the record to be verified.

[0039] After completing the continuous observation window processing, inter-frame displacement verification is first performed. The inter-frame displacement verification is limited to the following: within the continuous observation window, the center positions of the candidate regions of adjacent active infrared interactive candidate records are extracted one by one in the order of acquisition sequence number, and the corresponding local response unit numbers are extracted. The changes in the center positions of adjacent candidate regions are compared one by one to see if they are within the pre-locked displacement boundaries under the same rule version. At the same time, it is checked whether the corresponding local response unit numbers of adjacent candidate records are the same or adjacent. Only when both of the above two conditions are met is the adjacent record pair determined to pass the inter-frame displacement verification. When all adjacent record pairs in the continuous observation window pass, the candidate record corresponding to the end point of the continuous observation window is retained as a displacement continuous candidate. When any adjacent record pair fails, the candidate record corresponding to the end point of the continuous observation window is marked as a displacement mismatch record and will not proceed to the subsequent confirmation in this step.

[0040] Subsequently, peak stability verification is performed on the continuous candidate displacements. Peak stability verification is limited to the following: for all candidate records corresponding to the continuous candidate displacement within the continuous observation window, the changes in the coverage area and the boundary position of the candidate region are compared pairwise according to the acquisition sequence number. In this step, the peak stability verification uses the relationship between the changes in the coverage area and the boundary position of the candidate region as a unified judgment criterion, and no other intensity criteria are used. When the change in the coverage area of ​​the candidate region between adjacent candidate records does not exceed the preset area change boundary under the same rule version, and the change in the boundary position of the candidate region does not exceed the preset boundary change boundary under the same rule version, the adjacent record pair is determined to pass the peak stability verification. When all adjacent record pairs within the continuous observation window pass, the candidate record corresponding to the end point of the continuous observation window is retained and enters the background reversal verification. When any adjacent record pair fails, the candidate record corresponding to the end point of the continuous observation window is marked as a peak instability record and is not entered into the subsequent confirmation in this step.

[0041] For candidates that pass the peak stability check, a background reversal check is then performed. The background reversal check is defined as follows: First, the boundary positions of the candidate regions corresponding to the local response unit where the current candidate is located and its adjacent local response units are extracted, and then compared one by one with the boundary position change range corresponding to the response fluctuation range of the corresponding local response unit in the projected change reference field. If the boundary position change of the current candidate falls into the change range corresponding to the response fluctuation range, it is recorded as a background change candidate. Then, the boundary position of the candidate region of the background change candidate in the continuous observation window is compared one by one with the boundary position of the candidate region of the previous candidate record according to the acquisition sequence number to verify whether it forms a continuous boundary continuation chain. The continuous boundary continuation chain is defined as the continuous relationship in which the boundary positions of the candidate regions of adjacent candidate records in the continuous observation window change continuously in the same direction without boundary interruption. When the current candidate simultaneously meets the condition of falling into the change range corresponding to the response fluctuation range and cannot form a continuous boundary continuation chain, the candidate record corresponding to the end point of the continuous observation window is marked as a background reversal record and is not used as a valid interaction target. If neither of the above two conditions is met, it is determined that it passes the background reversal check.

[0042] A valid interactive target is confirmed only when the candidate record corresponding to the end point of the continuous observation window passes the inter-frame shift check, peak stability check, and background inversion check simultaneously.

[0043] Valid interactive targets are defined as on-site confirmed objects that are candidate records corresponding to the endpoint of the continuous observation window after being jointly confirmed by continuity, concentration, and background rejection. After confirmation, each target is written with the target interaction surface number, the start and end acquisition sequence number of the continuous observation window, the corresponding local response unit number, the valid target boundary position, the valid target center position, the confirmation status, the rule version identifier, the source candidate record identifier, and the confirmation basis marker, and stored in the local valid target record area of ​​the edge computing node for the next step to call. The confirmation status should at least be distinguished as confirmed, displacement mismatch, peak instability, background reversal, and pending verification. The confirmation basis marker should at least correspond to the inter-frame displacement verification conclusion, peak stability verification conclusion, and background reversal verification conclusion. Only one valid interactive target is confirmed for the same boundary continuous chain within the same continuous observation window. No duplicate confirmation is allowed. Old confirmation records should not be directly overwritten by the current confirmation result. Instead, the formation time and confirmation basis marker are retained by appending new confirmation records.

[0044] This step is applicable to scenarios where active infrared interactive candidate records are continuously confirmed during the operation of immersive projection platforms, interactive screens on facades, and multi-material non-touch interactive surfaces. It is not applicable to situations where the previous record has not been re-established after the lens mounting position has changed, the candidate record in the continuous observation window is severely broken, or the reference field fails due to projection changes.

[0045] Preferably, in the implementation of the display stand with a circular imaging area of ​​1.8 meters, the length of the continuous observation window can be set to 3 to 8 frames. The displacement boundary corresponds to the continuous movement range of 1 to 3 adjacent local response units. The area change boundary and the boundary change boundary are pre-locked according to the current acquisition rhythm and the size of the interaction surface. In the continuous 300 frame sampling, the consistency rate between the effective interaction target confirmation result and the actual continuous action trajectory of the interaction component can be no less than 95%. Without changing the technical concept of "performing inter-frame displacement verification, peak stability verification, and background reversal verification on active infrared interaction candidate records according to the continuous observation window to confirm the effective interaction target", the length of the continuous observation window and each verification boundary can be adjusted accordingly with the size of the target interaction surface, the acquisition rhythm, and the range of interaction actions.

[0046] S4. Extract the core response area and transition response area of ​​the effective interaction target, and generate the centroid correction value. The specific implementation is as follows: After confirming a valid interaction target in the previous step, the edge computing node immediately retrieves the current infrared image frame, valid target boundary position, valid target center position, corresponding local response unit number, and rule version identifier corresponding to the valid interaction target. It then performs core response area and transition response area extraction under the same target interaction surface, same lens mounting position, and same rule version. This process is automatically completed by the edge computing node and operates during the continuous acquisition period when the valid interaction target has been confirmed and is not invalid. The retrieval source is the local valid target record area and the current infrared image frame record area. During retrieval, the current infrared image frame is first aligned with the valid interaction target record according to the acquisition sequence number. Then, the consistency of the target interaction surface number, lens mounting position, and rule version identifier is checked. Only missing markers are allowed to be supplemented; the valid target boundary position, valid target center position, corresponding local response unit number, and rule version identifier are not supplemented. If there are discrepancies in acquisition sequence numbers, target interaction surface numbers, rule versions, or valid target boundary positions, the current round of extraction for the valid interaction target is stopped, and the process is transferred to the pending verification record.

[0047] After completing the recording and organization, the core response area of ​​the effective interactive target is extracted first. The core response area is defined as a continuous internal area within the boundary of the effective target, continuously connected to the center of the effective target, and maintaining a higher position than its surrounding adjacent areas in the current infrared image frame. During extraction, the boundary of the effective target is used as the outer boundary, and the center of the effective target is used as the starting reference. The infrared response distribution is checked along the continuous connection direction within the pixel range covered by the corresponding local response unit number. First, the high response area continuously connected to the center of the effective target is extracted, and then the fragmented edge parts near the boundary of the effective target and the discrete areas that are not continuously connected to the center of the effective target are excluded. When there are multiple separated areas, only the area continuously connected to the center of the effective target is retained as the core response area.

[0048] Subsequently, the transition response region of the effective interactive target is extracted. The transition response region is defined as the peripheral region located outside the core response region, continuously connected to the core response region, and whose infrared response is lower than the corresponding region of the core response region under the same rule version. During extraction, starting from the outer boundary of the core response region, the peripheral response distribution in the current infrared image frame is checked layer by layer along the effective target boundary position. Only the peripheral region that is continuously connected to the core response region, does not exceed the effective target boundary position, and has an infrared response lower than the corresponding region of the core response region is retained. Discontinuous regions, discrete regions, regions that exceed the effective target boundary position, and regions that have been identified as background inversion sources in previous steps are excluded. When the peripheral region does not show a reverse enhancement higher than the corresponding region of the core response region along the direction of the outer boundary of the core response region toward the effective target boundary position, the peripheral region is retained as the transition response region.

[0049] After extracting the core response area and transition response area, a centroid correction value is generated. The centroid correction value is defined as the position offset record used when correcting the current position of the effective target center position based on the relative distribution relationship between the core response area and the transition response area in the current infrared image frame. The formation order is as follows: First, the main positioning center is determined by the center position of the range surrounded by the core response area boundary. The main positioning center is the intermediate positioning position used when generating the centroid correction value. Then, the main distribution direction of the transition response area relative to the main positioning center is determined. The main distribution direction is the direction with the largest coverage of the transition response area relative to the main positioning center. When the coverage of the transition response area in the main distribution direction exceeds the preset correction trigger boundary under the same rule version, the centroid correction magnitude is determined based on the distance relationship between the outer boundary of the core response area and the outer boundary of the transition response area in the same direction, and the main distribution direction is determined as the centroid correction direction. When the transition response area is evenly distributed around the core response area and the difference in coverage in each direction does not exceed the preset difference boundary, the centroid correction value takes zero offset. When there is a slight imbalance in the transition response area but it does not reach the preset correction trigger boundary, the centroid correction value takes the preset minimum offset.

[0050] After generating the centroid correction value, write the target interaction surface number, acquisition sequence number, corresponding local response unit number, core response area boundary position, transition response area boundary position, effective target center position, main positioning center position, centroid correction direction, centroid correction magnitude, rule version identifier, source effective target identifier, and extraction basis marker item by item, and store them in the local correction record area of ​​the edge computing node for the next step to call. Among them, the extraction basis marker corresponds to at least the core response area extraction passed, the transition response area extraction passed, and the centroid correction value has been generated. Only one centroid correction value record is generated for the same effective interaction target within the same acquisition sequence number. It is not written repeatedly. The old record cannot be directly overwritten by the current record. Instead, the formation time and extraction basis marker are retained by appending new records.

[0051] This step is applicable to scenarios where immersive projection platforms, interactive wall screens, and multi-material non-touch interactive surfaces are used to center confirmed valid interactive targets during operation. It is not applicable to situations where the boundary position of the valid target is missing, the current infrared image frame cannot be aligned with the valid target record, or the valid interactive target has entered an invalid state.

[0052] Preferably, in the implementation of the display stand with a 1.8-meter circular imaging area, the core response area can be kept within a continuous internal area inside the effective target boundary position, and the transition response area can be kept as a continuous peripheral area located outside the core response area and lower than the corresponding area of ​​the core response area. When sampling for 300 consecutive frames, the center position after correction based on the centroid correction amount can maintain a more stable landing point consistency relative to the uncorrected center position. Without changing the technical concept of "extracting the core response area and transition response area of ​​the effective interactive target and generating the centroid correction amount", the boundary extraction aperture, correction trigger boundary and preset difference boundary of the core response area and transition response area can be adjusted accordingly based on the target interactive surface material, acquisition rhythm and interactive component action form.

[0053] S5. Based on the centroid correction, the projection change reference field, and the preset mapping matrix, generate the target mapping coordinates. The specific implementation is as follows: After generating the centroid correction value in the previous step, the edge computing node immediately retrieves the target interaction surface number, acquisition sequence number, corresponding local response unit number, effective target center position, main positioning center position, centroid correction direction, centroid correction magnitude, projection change reference field version number, and preset mapping matrix version number corresponding to that centroid correction value. It then generates target mapping coordinates under the same target interaction surface, the same lens mounting position, and the same rule version. This process is automatically completed by the edge computing node, and the effective time period is the continuous acquisition period during which the effective interaction target has been confirmed and the centroid correction value has been generated. The data sources are the local correction record area, the projection change reference field record area, and the preset mapping matrix record area. During retrieval, first align the centroid correction value record, projection change reference field record, and preset mapping matrix record according to the acquisition sequence number. Then verify the consistency of the target interaction surface number, lens mounting position association record, rule version identifier, projection change reference field version number, and preset mapping matrix version number. Only missing markers are allowed to be filled in; centroid correction direction, centroid correction magnitude, projection change reference field version number, and preset mapping matrix version number are not allowed to be filled in. If there are any issues such as mismatched acquisition sequence numbers, inconsistent target interaction surface numbers, inconsistent rule versions, missing projection change reference field, missing preset mapping matrix, or missing centroid correction value, stop the current round of coordinate generation for the current valid interaction target and transfer to the record to be checked.

[0054] After completing the recording and organization, the correction positioning position is first determined based on the centroid correction amount. In this step, the centroid correction amount is used as the position offset for correcting the current position of the effective target center position. Specifically, based on the effective target center position, the corresponding centroid correction amplitude is applied according to the centroid correction direction to obtain the correction positioning position uniquely corresponding to the current effective interactive target. When the centroid correction amount is zero offset, the correction positioning position remains consistent with the effective target center position. When the centroid correction amount is the preset minimum offset or has formed an actual offset, the correction positioning position is adjusted synchronously with the centroid correction direction and centroid correction amplitude. The correction positioning position is always limited to within the effective imaging boundary of the target interaction surface. When the centroid correction amount exceeds the effective imaging boundary of the target interaction surface, the nearest effective position within the effective imaging boundary is used as the correction positioning position.

[0055] Subsequently, pre-mapping constraint verification is performed based on the projection change reference field. In this step, the projection change reference field is used as a field reference record of the sensitivity of different local response units of the current target interaction surface to projection changes. During verification, the corresponding local response unit to which the corrected positioning position belongs is first determined, and then the stable response range boundary of the corresponding local response unit and adjacent local response units in the projection change reference field is retrieved. The stable response range boundary is the position boundary record of the corresponding local response unit in the projection change reference field that is allowed to be used as a stable mapping benchmark under this deployment. When the corrected positioning position does not cross the stable response range boundary, the corrected positioning position is directly entered into the mapping. When the corrected positioning position crosses the stable response range boundary, the corrected positioning position is intercepted by the stable response range boundary so that the verified corrected positioning position does not cross the stable response range boundary and does not enter the high fluctuation area corresponding to the response fluctuation interval. The high fluctuation area is the area record in the projection change reference field where the response fluctuation interval reaches the preset high fluctuation caliber under the current rule version.

[0056] After the aforementioned verification is completed, target mapping coordinates are generated based on a preset mapping matrix. The preset mapping matrix is ​​limited to a mapping record established and pre-locked based on the current target interaction surface position record and the output coordinate plane position record after this deployment is completed. It is used to characterize the correspondence between the current target interaction surface position and the output coordinate plane position used for subsequent interaction control, and is locked under the current rule version. During generation, the corrected positioning position verified by the projection change reference field constraint is used as the unique mapping reference. The preset mapping matrix record matching the current target interaction surface number is retrieved, and the corresponding position of the corrected positioning position in the output coordinate plane is determined based on the preset mapping matrix record, forming the target mapping coordinates uniquely corresponding to the current acquisition sequence number. The target mapping coordinates are limited to the output coordinate plane position coordinate record obtained after centroid correction, verification by the projection change reference field constraint, and transformation by the preset mapping matrix. The uncorrected valid target center position is not directly used as the mapping basis, nor is the corrected positioning position that has not completed constraint verification used as the mapping basis.

[0057] After generating the target mapping coordinates, the target interaction surface number, acquisition sequence number, corresponding local response unit number, effective target center position, corrected positioning position, projection change reference field version number, preset mapping matrix version number, target mapping coordinates, rule version identifier, source centroid correction amount identifier, and mapping basis mark are written item by item and stored in the local mapping record area of ​​the edge computing node for the next step to call. Among them, the mapping basis mark corresponds to at least the centroid correction amount has been adopted, the projection change reference field constraint has been verified, and the preset mapping matrix has been called. Only one target mapping coordinate record is generated for the same effective interaction target within the same acquisition sequence number. It is not written repeatedly. The old record shall not be directly overwritten by the current record. Instead, the formation time and mapping basis mark are retained by appending new records.

[0058] This step applies to scenarios where confirmed and corrected valid interactive targets are converted into output coordinates during the runtime of immersive projection platforms, interactive facades, and multi-material non-touch interactive surfaces. It does not apply to situations where the projection change reference field fails, the preset mapping matrix fails, the target interactive surface number does not match, or the centroid correction amount is missing.

[0059] Preferably, in the implementation of the display stand with a 1.8-meter circular imaging area, the edge computing node can complete the correction and positioning verification and target mapping coordinate generation within a single frame period. In a continuous 300-frame sampling inspection, the target mapping coordinates generated by the centroid correction amount, the projection change reference field and the preset mapping matrix can maintain a higher landing point stability and consistency than the uncorrected direct mapping method. Without changing the technical concept of "generating target mapping coordinates based on centroid correction amount, projection change reference field and preset mapping matrix", the recording aperture, stable response range boundary and high fluctuation area aperture of the preset mapping matrix can be adjusted accordingly with the target interaction surface size, surface material and output coordinate plane arrangement.

[0060] S6. Perform rhythmic constraint output on the target mapped coordinates to generate structured tracking results, and update the projected change reference field according to local perturbation changes. The specific implementation is as follows: After generating the target mapping coordinates in the previous step, the edge computing node immediately retrieves the target interaction surface number, acquisition sequence number, corresponding local response unit number, target mapping coordinates, projection change reference field version number, rule version identifier, and source centroid correction value identifier corresponding to the target mapping coordinates. Under the same target interaction surface, the same lens mounting position, and the same rule version, it performs rhythmic constraint output on the target mapping coordinates. This execution is automatically completed by the edge computing node, and the effective period is the continuous acquisition period during which the effective interaction target exists and the target mapping coordinates are continuously generated. The source of the data is the local mapping record area and the projection change reference field. Recording area; during retrieval, first align the target mapping coordinate record and the projection change reference field record according to the acquisition sequence number, then verify the consistency of the target interaction surface number, the corresponding local response unit number, the projection change reference field version number, and the rule version identifier. Only missing markers are allowed to be filled in; target mapping coordinates, corresponding local response unit numbers, projection change reference field version numbers, and rule version identifiers are not allowed to be filled in. If there are cases where the acquisition sequence number cannot be matched, the target interaction surface number is inconsistent, the rule version is inconsistent, the target mapping coordinates are missing, or the projection change reference field is missing, stop the rhythm constraint output corresponding to the current acquisition sequence number and switch to the record to be checked.

[0061] After completing the record organization, the target mapping coordinates corresponding to the current collection sequence number are first used as the current output candidate, and the current output position that was not transferred to the pending verification state before the current collection sequence number is retrieved as the previous valid output position; when there is no previous valid output position, the current output candidate is directly used as the current output position.

[0062] When a previous valid output position exists, rhythmic constraint output is applied to the target mapping coordinates. The rhythmic constraint output is limited to: using the positional change relationship between the current output candidate and the previous valid output position as the verification object, determining whether the positional change satisfies the pre-locked output rhythm boundary under the same rule version. The output rhythm boundary is the allowable output change range jointly determined by the interaction speed of the current target interaction surface, the acquisition rhythm, and the output coordinate plane refresh rhythm; when the current position change does not exceed the output rhythm boundary, the current output candidate is directly used as the current output position; when the current position change exceeds the output rhythm boundary, the current output candidate is retained as the current output position; when the current position change exceeds the output rhythm boundary, the current output candidate is used as the current output position. When the rhythm boundary is reached, the current output position is restricted to the allowable output range with the previous valid output position as the reference and the output rhythm boundary as the outer boundary, so that the current output position does not undergo a leap in output beyond the output rhythm boundary. For example, in a projection table scene, when a visitor holds an interactive device and undergoes a large displacement change between consecutive acquisition sequences, the target mapping coordinate may show an instantaneous increase in offset between adjacent acquisition sequences. At this time, the rhythm constraint output restricts the current output position to the allowable output change range of the current refresh cycle, avoiding abrupt changes in the projection feedback position on the screen that exceed the output rhythm boundary.

[0063] After completing the rhythm constraint output, a structured tracking result is generated. This structured tracking result is limited to a unified record of the target mapping coordinates and their associated field states for the target whose rhythm constraint output has been completed under the current acquisition sequence number. During its generation, the following information is written item by item: target interaction surface number, acquisition sequence number, corresponding local response unit number, current output position, previous valid output position, rhythm constraint state, target mapping coordinates, projection change reference field version number, rule version identifier, source target mapping coordinate identifier, and output basis marker. This information is then stored in the local tracking result record area of ​​the edge computing node. The target mapping coordinates are those before the execution of the rhythm constraint output. The current output position is the post-constraint position record after the execution of rhythm constraint output. The previous valid output position is the most recent post-constraint position record before the current acquisition sequence number that has not yet entered the pending verification state. The rhythm constraint state must at least distinguish between direct output, post-constraint output, and pending verification. The output basis marker must at least correspond to the target mapping coordinates being called, the rhythm constraint output being executed, and the current output position being determined. Within the same acquisition sequence number, only one structured tracking result is generated for the same target interaction surface. It is not written repeatedly. Old records cannot be directly overwritten by the current record. Instead, the formation time and output basis marker are retained by appending new records.

[0064] Subsequently, the projected change reference field is updated according to local disturbance changes. Local disturbance changes are limited to the response change records of the corresponding local response unit and its adjacent local response units associated with the current output position within a preset update interval, relative to the deviation relationship formed by the response fluctuation interval records of the corresponding local response units in the existing projected change reference field. During the update, the corresponding local response unit number associated with the current output position corresponding to the structured tracking result is used as the main factor. The response change records of the corresponding local response unit and its adjacent local response units within the current acquisition sequence number to the preset update interval are continuously extracted. Then, they are compared item by item with the response fluctuation interval records of the corresponding local response units in the existing projected change reference field to verify whether the deviation has continuously reached the preset update trigger under the same rule version within the preset update interval. The system identifies the boundary; simultaneously, it verifies whether the local response unit change sequence corresponding to the deviation is continuously distributed along the local response unit change sequence corresponding to the current output position. If the local response unit change sequence corresponding to the deviation is not continuously distributed along the local response unit change sequence corresponding to the current output position, and continuously reaches the preset update trigger boundary within the preset update interval, the local disturbance change is written into the new fluctuation record of the local response unit corresponding to the projection change reference field to update the projection change reference field. If the deviation is only a single frame fluctuation, does not continuously reach the preset update trigger boundary, or its corresponding local response unit change sequence is continuously distributed along the local response unit change sequence corresponding to the current output position, the projection change reference field is not updated, thereby avoiding the miswriting of the real interactive target trajectory as background change.

[0065] After the update is completed, the target interaction surface number, acquisition sequence number, local response unit number corresponding to the local disturbance change, projection change reference field update status, version number before update, version number after update, source structured tracking result identifier, and newly added fluctuation record identifier are written item by item. The updated projection change reference field record is then written back to the local reference field record area for subsequent acquisition sequence numbers to continue to call. The newly added fluctuation record includes at least the corresponding local response unit number, deviation start and end acquisition sequence number, response fluctuation interval identifier before update, newly added fluctuation identifier after update, and update basis status.

[0066] This step is applicable to scenarios where immersive projection platforms, interactive facades, and multi-material non-touch interactive surfaces are used to smoothly constrain continuous output positions and incrementally correct background change reference records during operation. It is not applicable to situations where the target interactive surface numbers do not match, the output coordinates are severely interrupted, or the projection change reference field has become invalid.

[0067] Preferably, in the implementation on the display stand with a circular imaging area of ​​1.8 meters, rhythmic constraint output and structured tracking results can be completed in a single frame cycle. Local disturbance changes can be determined within a preset update interval of 3 to 10 consecutive frames. In a continuous 300-frame sampling inspection, the current output position after using rhythmic constraint output can maintain a higher consistency of landing point continuity. Without changing the technical concept of "performing rhythmic constraint output on the target mapping coordinates, generating structured tracking results, and updating the projection change reference field according to local disturbance changes", the output rhythm boundary, update trigger boundary, and preset update interval can be adjusted accordingly with the target interaction surface size, acquisition rhythm, and output coordinate plane refresh rhythm.

[0068] In the operational scenario shown in this embodiment: This method is deployed on an immersive circular projection interactive booth in a science and technology exhibition hall. The target interactive surface of the booth is a circular imaging area with a diameter of 1.8 meters. The surface is a coated imaging surface. Edge computing nodes are installed in the equipment compartment below the booth. An infrared image sensor is fixedly installed in a hidden observation position 2.6 meters directly above the target interactive surface, ensuring that the current field of view of the infrared image sensor completely covers the effective imaging boundary of the target interactive surface. The projection equipment projects the exhibit images onto the target interactive surface using a top-mounted method. When the exhibits are in operation, they include static star map images, images with sudden changes in local brightness, images with flowing textures, and images with continuous motion.

[0069] After the on-site commissioning personnel complete the installation, the edge computing node is powered on and enters the calibration preparation period. First, under the conditions that the active infrared interactive target transmitter is turned off, no one enters the target interaction surface with an interactive component, the projection equipment maintains the baseline standby screen, and the on-site lighting maintains the standard lighting level, infrared reference frames are continuously acquired. Then, under the same lens installation position, the same exposure aperture, the same gain aperture, and the same rule version, various formal operation projection segments are played and projection change reference frames are continuously acquired. After the acquisition of the two types of reference frames is completed, the edge computing node is aligned according to the acquisition completion time, and the continuous occlusion frames caused by the commissioning personnel passing by for a short time and the local saturation frames caused by the switching of projection brightness are removed. Then, the continuity of the acquisition sequence number is checked. After confirming that there are no continuous missing numbers, 196 local response units are divided according to the effective imaging boundary of the target interaction surface according to the preset grid rules. The infrared reference frame observation results and projection change reference frame observation results are merged in each local response unit to form the basic infrared response range and response fluctuation range of each local response unit. The projection change reference field is generated and written to the local storage area of ​​the edge computing node. At the same time, the version number of the local response unit and the version number of the projection change reference field are submitted to the upper management terminal.

[0070] Once the exhibit is in full operation, visitors hold the interactive device close to the target interactive surface. The edge computing node continuously acquires the current infrared image frame. Within each current infrared image frame, infrared difference enhancement is performed according to the local response unit. First, the observation part outside the invalid imaging boundary is eliminated. Then, the current infrared observation value is compared with the basic infrared response range and response fluctuation range one by one to eliminate the influence of stable background and projection changes. The remaining infrared variation part is used as the candidate formation object. For example, when a visitor moves the interactive device to the upper right quadrant of the target interactive surface, the edge computing node obtains a continuous high response area in three adjacent local response units and merges them into a continuous coverage area in a single frame. After verifying that the continuous coverage area is within the effective imaging boundary, does not cross non-adjacent local response units, and can form a unique candidate area boundary, an active infrared interactive candidate record is formed and written into the candidate record area.

[0071] Subsequently, the edge computing node takes 5 frames backward from the acquisition sequence number around the current candidate record to be confirmed to form a continuous observation window. It performs inter-frame displacement verification, peak stability verification, and background inversion verification on the active infrared interaction candidate record. When the center position of the candidate region remains within the preset displacement boundary within the continuous observation window, the changes in the coverage area and boundary position of the candidate region do not exceed the preset area change boundary and the preset boundary change boundary, and the boundary position of the candidate region passes through the edge of the local bright video but does not fall into the boundary position change range corresponding to the response fluctuation interval, or although it falls into the range, it still forms a continuous boundary chain, the edge computing node confirms the candidate record corresponding to the end of the continuous observation window as a valid interaction target and writes the valid target boundary position, valid target center position, confirmation status, and confirmation basis marker.

[0072] For confirmed valid interactive targets, the edge computing node continues to retrieve the current infrared image frame. Within the boundary of the valid target, it extracts a continuous high-response internal region that is continuously connected to the center of the valid target as the core response area. Then, it extracts a continuous peripheral region along the outer boundary of the core response area towards the boundary of the valid target, where the infrared response is lower than that of the corresponding region of the core response area, as the transition response area. For example, when the audience slightly tilts the interactive object closer to the target interactive surface, the edge computing node recognizes that the core response area is slightly narrower on the left side of the target, while the transition response area on the right side has a larger coverage area. Thus, it determines that the main distribution direction of the transition response area relative to the main positioning center is on the right side, and generates a centroid correction amount pointing to the right based on the distance relationship between the outer boundary of the core response area and the outer boundary of the transition response area.

[0073] Next, the edge computing node corrects the effective target center position based on the centroid correction amount to obtain the corrected positioning position. If the corrected positioning position is close to the high fluctuation projection edge of a certain local response unit, the corrected positioning position is then intercepted based on the stable response range boundary in the projection change reference field to prevent the position from entering the high fluctuation region.

[0074] After the constraint verification is completed, the edge computing node calls the preset mapping matrix that matches the current target interaction surface number to convert the corrected positioning position into the target mapping coordinates in the output coordinate plane.

[0075] Taking the aforementioned booth as an example, when the audience holds the interactive component and slides it clockwise along the outer edge of the truncated cone, the edge computing node continuously generates a set of target mapping coordinates that change along the circumference.

[0076] To avoid output jumps caused by audience hand tremors, changes in projection brightness, or instantaneous shifts in a single frame, the edge computing node further uses the target mapping coordinates corresponding to the current acquisition sequence number as the current output candidate and retrieves the previous valid output position to perform rhythm constraint output. If the change in the current position does not exceed the output rhythm boundary, it is output directly. If it exceeds the output rhythm boundary, the current output position is restricted to the allowable output range based on the previous valid output position, and then a structured tracking result containing the current output position, the previous valid output position, the rhythm constraint state, and the target mapping coordinates is generated.

[0077] For example, when a viewer quickly drags the interactive element from the right side of the circular platform to the center, the original target mapping coordinates show a large offset between adjacent acquisition numbers. The edge computing node uses rhythm constraint output to smoothly limit the current position within the allowable range of the refresh cycle, so that the projected feedback light point moves continuously on the screen instead of jumping and flashing.

[0078] Meanwhile, the edge computing node will also extract response change records within a preset update interval of 5 consecutive frames based on the corresponding local response unit and its adjacent local response units associated with the current output position, and compare them item by item with the response fluctuation interval records in the existing projection change reference field. When a local area continuously deviates from the original response fluctuation interval due to the environmental reflector temporarily added on site, and the change sequence of the local response unit corresponding to the deviation is not continuously distributed along the change sequence of the local response unit corresponding to the current output position, the edge computing node writes the local disturbance change into the new fluctuation record, updates the projection change reference field, and writes back to the local reference field record area.

[0079] Through continuous operation, the system can continuously output stable target mapping coordinates and structured tracking results under actual conditions such as formal playback of exhibits, continuous audience interaction, local projection brightness changes, and minor disturbances in the on-site environment. It also performs incremental corrections to the projection change reference field, thereby maintaining the stability and traceability of the active infrared interactive target tracking results.

[0080] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0081] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0082] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0083] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0085] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0087] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0089] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for active infrared interactive target tracking based on edge computing, characterized in that, include: S1. Acquire infrared reference frames and projection change reference frames of the target interaction surface, divide the local response units, and generate the projection change reference field. S2. Acquire the current infrared image frame, perform infrared difference enhancement according to the local response unit, and form an active infrared interactive candidate record; S3. Perform inter-frame shift verification, peak stability verification, and background inversion verification on active infrared interaction candidate records according to continuous observation windows to confirm valid interaction targets; S4. Extract the core response area and transition response area of ​​the effective interaction target, and generate the centroid correction value; S5. Generate target mapping coordinates based on the centroid correction amount, projection change reference field, and preset mapping matrix; S6. Perform rhythmic constraint output on the target mapping coordinates to generate structured tracking results, and update the projected change reference field according to local perturbation changes.

2. The active infrared interactive target tracking method based on edge computing according to claim 1, characterized in that, S1 includes: Infrared reference frames and projection change reference frames are acquired under the same target interaction surface, the same lens mounting position, the same exposure aperture, the same gain aperture, and the same rule version. Then, local response units are divided based on the effective imaging boundary of the target interaction surface; Then, for each local response unit, the infrared reference frame observation results are collected to form the basic infrared response lower limit and basic infrared response upper limit, and the projection change reference frame observation results are collected to form the response fluctuation lower limit and response fluctuation upper limit, so as to generate the projection change reference field.

3. The active infrared interactive target tracking method based on edge computing according to claim 1, characterized in that, S2 include: Under the same lens mounting position, the same target interaction surface, and the same rule version, continuously acquire the current infrared image frame; Based on the local response unit, the observation part outside the invalid imaging boundary is excluded within the corresponding pixel range, and the remaining current infrared observation value is compared with the basic infrared response range determined by the basic infrared response lower limit and the basic infrared response upper limit, and the response fluctuation range determined by the response fluctuation lower limit and the response fluctuation upper limit in turn, and the infrared change part is retained as a candidate formation object. Candidate objects are then merged within the same local response unit to form a continuous coverage area, and the continuous coverage area is verified according to the closed candidate formation caliber to form an active infrared interactive candidate record.

4. The active infrared interactive target tracking method based on edge computing according to claim 1, characterized in that, S3 includes: Active infrared interactive candidate records are retrieved continuously according to the acquisition sequence number, and a continuous observation window is formed with the acquisition sequence number of the current active infrared interactive candidate record as the window endpoint. Within a continuous observation window, first perform inter-frame shift verification, then perform peak stability verification, and retain the candidate records corresponding to the end point of the continuous observation window that pass the peak stability verification to enter the background inversion verification.

5. The active infrared interactive target tracking method based on edge computing according to claim 4, characterized in that, S3 also includes: Perform background inversion verification on the candidate records corresponding to the endpoints of the continuous observation windows that have passed the peak stability verification; When the candidate record corresponding to the end of the continuous observation window falls into the boundary position change range of the response fluctuation interval and cannot form a continuous boundary continuation chain, it is marked as a background reversal record. Candidate records corresponding to the endpoints of consecutive observation windows that are not marked as background inversion records are used as valid interaction target confirmation objects.

6. The active infrared interactive target tracking method based on edge computing according to claim 1, characterized in that, S4 includes: Under the same target interaction surface, the same lens mounting position, and the same rule version, retrieve the current infrared image frame, the effective target boundary position, the effective target center position, and the corresponding local response unit number corresponding to the effective interaction target; Then, within the pixel range covered by the corresponding local response unit number, extract the high-response continuous internal region that is continuously connected to the effective target center position as the core response region, and extract the continuous peripheral region located outside the core response region that is continuously connected to the core response region and lower than the corresponding region of the core response region as the transition response region. The main positioning center is then determined by the center of the range enclosed by the core response area boundary, and the centroid correction amount is generated based on the main distribution direction of the transition response area relative to the main positioning center.

7. The active infrared interactive target tracking method based on edge computing according to claim 1, characterized in that, S5 include: Based on the effective target center position, apply the centroid correction amplitude in the centroid correction direction to obtain the corrected positioning position; Next, determine the corresponding local response unit to which the corrected positioning position belongs, and perform pre-mapping constraint verification on the corrected positioning position based on the stable response range boundary of the corresponding local response unit determined by the projection change reference field.

8. The active infrared interactive target tracking method based on edge computing according to claim 7, characterized in that, S5 also includes: Using the corrected positioning position verified by pre-mapping constraints as the unique mapping reference, the preset mapping matrix record matching the target interaction surface number is called to determine the corresponding position of the corrected positioning position in the output coordinate plane and generate the target mapping coordinates.

9. The active infrared interactive target tracking method based on edge computing according to claim 1, characterized in that, S6 include: The target mapping coordinates corresponding to the current acquisition sequence number are used as the current output candidate, and the current output position that was not transferred to the pending verification state before the current acquisition sequence number is retrieved as the previous valid output position. When the current position changes beyond the output rhythm boundary, the current output position is restricted to the allowable output range with the previous valid output position as the reference and the output rhythm boundary as the outer boundary. The structured tracking results are then regenerated, and the projected reference field is updated according to the local perturbation changes.

Images