An image recognition-based firework detection system for charging area of two-wheeled electric vehicle

Abstract

The present application relates to the technical field of electric vehicle charging area fire detection, more specifically, the present application relates to a two-wheeled electric vehicle charging area fire detection system based on image recognition, which collects continuous image frames of the charging area, identifies the vehicle, charging socket, charger and charging cable area, generates an effective charging link snapshot with link identification when the charging cable ends and the vehicle end and power supply end maintain stable and adjacent across frames, identifies the smoke and flame candidate area and generates a smoke and flame candidate trajectory, determines the candidate source end in the initial appearance window, judges the source end anchoring relationship with the risk anchoring area, accumulates the smoke and flame evolution confidence with effective correlation frames, accumulates the frame attenuation or interruption, and outputs the smoke and flame risk judgment result bound with link identification.

Description

Technical Field

[0001] This invention relates to the field of smoke and fire detection technology in electric vehicle charging areas, and more specifically, to a smoke and fire detection system for two-wheeled electric vehicle charging areas based on image recognition. Background Technology

[0002] With the increasing use of two-wheeled electric vehicles in residential communities, businesses, commercial spaces, and underground parking garages, the number of centralized charging areas is gradually increasing. The charging process for two-wheeled electric vehicles typically involves multiple components, including the vehicle itself, the battery mounting area, the charger, charging cables, and charging sockets, forming an actual charging connection. If the battery, charger, cables, or socket experience overheating, short circuits, or poor contact, the initial symptoms often manifest as localized smoke, bright flames, or small-scale fires, which can spread to adjacent vehicles within a short period.

[0003] Current safety monitoring methods for charging areas mainly include smoke detectors, temperature sensors, manual inspections, and video surveillance. Smoke and temperature sensors are significantly affected by their installation location, airflow, ceiling height, and obstructions. They typically require smoke or temperature rise to reach the vicinity of the sensor before triggering an alarm, making them insufficient for directly responding to early, localized anomalies near chargers, cable ends, or vehicle batteries. Manual inspections rely on on-duty personnel and are difficult to cover nighttime, unattended, or multi-point, dispersed charging scenarios.

[0004] Image recognition technology has been used for smoke and flame detection, capturing surveillance footage through cameras and identifying suspected smoke or flame targets within the images. While this approach is suitable for general fire monitoring scenarios, it still has limitations in charging areas for two-wheeled electric vehicles. Common interferences in charging areas include vehicle lights, streetlights, charging indicator lights, reflections from metal frames, obstructions from people, vehicle movement, and changes in ambient light. Relying solely on smoke color, flame brightness, or continuous frame changes for judgment can easily lead to misidentifying non-fire targets as smoke or flame targets.

[0005] On the other hand, existing video smoke detection methods typically focus on the presence of smoke or flames in the frame, but rarely determine whether a suspected smoke target originates from a specific charging link. For example, suspected smoke may appear in the frame, but it could be located in the background or near a non-charging object; a suspected bright target may be close to the vehicle, but it could simply be a fixed indicator light or a reflective point. Without distinguishing the actual connections between the vehicle, charging cable, charging socket, and charger, it is difficult to determine whether a smoke candidate is related to the same charging process.

[0006] Furthermore, early smoke or fire points are typically highly localized, and their risk assessment relies heavily on whether the source is close to the vehicle battery area, the end of the charging cable, the charger, or the socket. Existing methods often use the target bounding box, target center, or overall frame confidence as the basis for judgment, lacking continuous correlation assessment of the initial source of the smoke and fire candidate trajectory. When the candidate target deviates from the charging link in subsequent frames, or exhibits interference characteristics such as stable brightness or stable outline, there is also a lack of corresponding reverse suppression mechanisms, making it difficult for alarm results to accurately correspond to the actual charging risk location.

[0007] Therefore, we propose an image recognition-based smoke and fire detection system for the charging area of ​​two-wheeled electric vehicles to solve the above problems. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition, in order to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition, comprising an image acquisition module, a link snapshot generation module, a source end anchoring module, an evidence frame generation module, and a risk determination module; The image acquisition module is used to acquire continuous image frames of the charging area of ​​two-wheeled electric vehicles; The link snapshot generation module is used to identify the vehicle area, charging socket area, charger area and charging cable area in continuous image frames, and generate a valid charging link snapshot with a link identifier when the two ends of the charging cable area are respectively adjacent to the vehicle area, charging socket area or charger area and are stable across frames. The source-end anchoring module is used to identify smoke candidate regions and flame candidate regions in consecutive image frames, generate smoke and fire candidate trajectories, determine candidate source ends within the initial appearance window of the smoke and fire candidate trajectory, and determine whether the candidate source ends form a source-end anchoring relationship with the risk anchoring region in the same valid charging link snapshot. The evidence frame generation module is used to record image frames that form a source-end anchoring relationship as valid association frames, and image frames that do not maintain the source-end anchoring relationship and meet the preset interference characteristics as counter-evidence frames; The risk assessment module is used to accumulate the confidence level of fireworks evolution based on valid associated frames, and to generate a fireworks risk assessment result with a bound link identifier based on the accumulation of confidence level of fireworks evolution attenuated or interrupted by counter-evidence frames.

[0010] In a preferred embodiment, an effective charging link snapshot includes the same vehicle area, the vehicle end of the charging cable adjacent to the vehicle area, the charging socket area adjacent to the power supply end of the charging cable, and the charger area located between the vehicle area and the charging socket area.

[0011] In a preferred embodiment, the link snapshot generation module maintains, updates, or releases the link identifier based on the overlap between the vehicle area, charging cable area, charging socket area, and charger area in adjacent image frames, as well as the proximity of the ends of the charging cable.

[0012] In a preferred embodiment, the risk anchoring area includes at least two of the following: the vehicle battery associated area, the charging cable vehicle end adjacent area, the charging cable power supply end adjacent area, the charger area, and the charging socket area.

[0013] In a preferred embodiment, the source-end anchoring module determines the candidate source end based on the boundary of the smoke candidate region or the flame candidate region on the side closest to the risk anchoring region within the initial appearance window, the starting position of the continuous frame, or the position deduced from the expansion direction.

[0014] In a preferred embodiment, the evidence frame generation module records the corresponding image frame as a valid associated frame when the candidate source is located within the risk anchoring area or when the distance between the candidate source and the risk anchoring area meets a preset proximity condition.

[0015] In a preferred embodiment, the risk determination module accumulates smoke evolution confidence for smoke candidate trajectories in the effective associated frames based on area growth, edge diffusion, transparency changes, or center position migration, and accumulates flame evolution confidence for flame candidate trajectories in the effective associated frames based on brightness fluctuations, contour jumps, color changes, or area jumps.

[0016] In a preferred embodiment, the preset interference characteristics include a stable target position and no source-end anchoring relationship, stable target brightness and contour changes below preset change conditions, or a target trajectory continuously deviating from the corresponding effective charging link snapshot.

[0017] In a preferred embodiment, the system further includes an alarm output module, which outputs alarm information including link identifier, candidate source location, risk anchoring area, valid association frame, counter-evidence frame and fire evolution confidence level when generating the fire risk determination result.

[0018] The technical effects and advantages of this invention are as follows: 1. This invention focuses on detecting smoke and fire within the actual charging connection area of ​​two-wheeled electric vehicles, rather than relying solely on smoke and flame targets in the overall monitoring image as alarm criteria. The system first forms an effective charging link snapshot from the vehicle area, charging cable area, charging socket area, and charger area, and distinguishes different vehicles or different charging connection relationships through link identifiers. This allows subsequent smoke and fire detection to be mapped to specific charging links, avoiding the direct attribution of abnormal images in the background area or near unrelated vehicles to charging fires.

[0019] 2. This invention focuses on the initial source of candidate smoke and fire trajectories for risk assessment. By determining whether the candidate source establishes an anchoring relationship with the vehicle battery area, the area adjacent to the charging cable end, the charger area, or the charging socket area, it enables early smoke, small-scale fire points, and other local anomalies to be associated with the risk anchoring areas in the corresponding valid charging link snapshots. Compared to methods that only determine the target box position or identify confidence in a single frame, this processing is more suitable for scenarios where charging anomalies in two-wheeled electric vehicles typically begin near the battery, cable end, charger, or socket.

[0020] 3. This invention accumulates the confidence level of fireworks evolution through effective correlation frames and uses counter-evidence frames to attenuate or interrupt the confidence level, so that the fireworks confirmation process is simultaneously constrained by the charging link, the source position, and the evolution of consecutive frames. For common interference targets such as vehicle lights, charging indicator lights, metallic reflections, stable high-brightness spots, and short-term obstructions, if they do not form a source-end anchoring relationship with the risk anchoring area, or if they exhibit stable position, insufficient contour changes, or trajectory deviation from the effective charging link in consecutive frames, they will not directly enter the effective fireworks accumulation process.

[0021] 4. The fire risk results output by this invention are bound to the link identifier, candidate source location, risk anchoring area, valid association frame and counter-evidence frame, so that the alarm result not only indicates the existence of fire risk, but also points to the corresponding vehicle, cable end, charger or socket area. This allows the alarm information to carry the corresponding link identifier, candidate source location and risk anchoring area, and retains image evidence for subsequent review, handling and event tracing. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram illustrating the composition and recording content of an effective charging link snapshot in this invention; Figure 3 This is a schematic diagram illustrating the state of link identifier maintenance, updating, and release in consecutive image frames in this invention; Figure 4 This is a schematic diagram illustrating the division of risk anchoring areas under the same effective charging link snapshot in this invention; Figure 5 This is a schematic diagram illustrating the determination of candidate source ends within the initial appearance window of the fireworks candidate trajectory in this invention; Figure 6 This is a schematic diagram illustrating the generation of a valid associated frame set based on the source-end anchoring relationship in this invention; Figure 7 This is a schematic diagram illustrating the cumulative confidence level of fireworks evolution based on effective associated frames in this invention. Figure 8This is a schematic diagram illustrating the attenuation or interruption of the confidence level of the fireworks evolution based on the counter-evidence frame in this invention; Figure 9 This is a schematic diagram of the smoke and fire risk assessment results and alarm output information in this invention. Detailed Implementation

[0023] 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.

[0024] A smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition. The overall system structure of this embodiment is as follows: Figure 1 As shown, it includes an image acquisition module, a link snapshot generation module, a source end anchoring module, an evidence frame generation module, and a risk determination module; Each module transmits data according to the processing order of continuous image frames. Specifically, the image acquisition module outputs continuous image frames, the link snapshot generation module forms a valid charging link snapshot based on the continuous image frames, the source end anchoring module performs source end judgment on the candidate trajectory of fireworks based on the valid charging link snapshot, the evidence frame generation module forms a valid correlation frame and a counter-evidence frame, and the risk judgment module outputs the fireworks risk judgment result based on the valid correlation frame and the counter-evidence frame.

[0025] The image acquisition module is used to acquire continuous image frames of the charging area of ​​two-wheeled electric vehicles; The continuous image frames are formed into a frame sequence according to the acquisition time. Each image frame has a frame number and acquisition time, and at least covers the vehicle parking position, the charging socket position, the charging cable passage position, and the charger placement position, which serve as the image data basis for subsequent cross-frame association judgment.

[0026] The link snapshot generation module is used to identify the vehicle area, charging socket area, charger area and charging cable area in continuous image frames, and generate a valid charging link snapshot with a link identifier when the two ends of the charging cable area are respectively adjacent to the vehicle area, charging socket area or charger area and are stable across frames. The link snapshot generation module can identify vehicle areas, charging socket areas, and charger areas using object detection models, semantic segmentation models, or contour feature-based image recognition methods. For the charging cable area, it can first extract long, thin linear targets from the image, and then filter them based on their color, width, direction, and proximity to the vehicle and charging socket areas. The recognition results should include at least the boundary location, center location, and image frame number of each target area.

[0027] The aforementioned target detection model or semantic segmentation model can be trained using pre-labeled image samples of the charging area of ​​two-wheeled electric vehicles. The labeled objects in the image samples include the vehicle area, charging socket area, charger area, and charging cable area. The output of the recognition model can be a target bounding box, a region mask, or contour coordinates. For the charging cable area, elongated linear areas that are adjacent to the vehicle area, charging socket area, or charger area are preferentially retained, while background linear targets unrelated to the charging link are removed.

[0028] The two ends of the charging cable area can be determined by the outline endpoints, skeleton endpoints, or linear area ends of the charging cable area; end proximity means that the vehicle end of the charging cable area and the vehicle area, or the power supply end of the charging cable area and the charging socket area or charger area, meet a preset distance condition or have a contact relationship. Cross-frame stability means that the above-mentioned end proximity relationship is maintained in multiple consecutive image frames, and the overlap relationship of the corresponding vehicle area, charging cable area, charging socket area, or charger area meets a preset stability condition.

[0029] The preset distance conditions can be set according to the image pixel distance, the size ratio of the target area's bounding box, or the actual calibrated distance; for example, if the distance between the end of the charging cable and the boundary of the vehicle area, the boundary of the charging socket area, or the boundary of the charger area is less than the preset proximity distance, it is considered that the end proximity is satisfied. The preset stability conditions may include: the end proximity relationship exists in multiple consecutive image frames, and the degree of overlap of the corresponding target areas in adjacent image frames reaches the preset requirement, or the target center position offset is less than the preset offset range.

[0030] When no actual distance calibration is performed, the preset proximity distance and preset offset range can be normalized according to the width of the outer frame of the vehicle area, the size of the outer frame of the charger area, or the size of the outer frame of the charging socket area in the image. The degree of overlap of the same target area in adjacent image frames can be represented by the area overlap ratio. When the area overlap ratio reaches a preset value, the target area is considered to remain stable in adjacent image frames.

[0031] A valid charging link snapshot includes the same vehicle area, the vehicle end of the charging cable adjacent to the vehicle area, the charging socket area adjacent to the power supply end of the charging cable, and the charger area located between the vehicle area and the charging socket area. The composition and recorded content of a valid charging link snapshot can be found in [link to relevant documentation]. Figure 2 The effective charging link snapshot records the link identifier, vehicle area location, charging cable vehicle end location, charging cable power supply end location, charging socket area location, charger area location, and corresponding image frame information. The same link identifier is used to indicate the same vehicle and the corresponding charging connection relationship.

[0032] The link snapshot generation module maintains, updates, or releases the link identifier based on the overlap between the vehicle area, charging cable area, charging socket area, and charger area in adjacent image frames, as well as the proximity of the ends of the charging cable.

[0033] The maintenance, update, and release status of the link identifier in consecutive image frames can be found in [reference]. Figure 3 The overlap relationship of regions can be determined by the overlap ratio, center position offset, or boundary position offset of similar target regions in adjacent image frames. For charging cable regions, it can also be determined by combining the cable skeleton orientation, end position offset, and linear region continuity. When similar target regions satisfy the region overlap relationship in adjacent image frames, they are regarded as cross-frame continuation of the same target region.

[0034] When the vehicle area, charging cable area, charging socket area, and charger area in adjacent image frames maintain the same area overlap, and the two ends of the charging cable still satisfy the end proximity relationship, the original link identifier is maintained; when the end proximity relationship changes or the corresponding area overlap relationship does not meet the stable condition, the link identifier is updated; when the proximity relationship between the end of the charging cable and the vehicle area, charging socket area, or charger area disappears in consecutive image frames, the corresponding link identifier is released.

[0035] The risk anchoring area includes at least two of the following: the vehicle battery associated area, the area near the vehicle end of the charging cable, the area near the power supply end of the charging cable, the charger area, and the charging socket area. For risk anchoring area delineation under the same valid charging link snapshot, please refer to Figure 4 The risk anchoring area is selected from the same valid charging link snapshot and bound to the link identifier of that valid charging link snapshot. For the same link identifier, the vehicle battery associated area is used to characterize the risk location on the vehicle side, the charging cable vehicle end adjacent area and the charging cable power supply end adjacent area are used to characterize the risk location at the cable connection end, and the charger area and charging socket area are used to characterize the risk location on the power supply side.

[0036] The vehicle battery associated area can be determined based on the vehicle's body posture, vehicle outline proportions, and preset battery installation location. When the specific battery outline cannot be identified, a local area in the vehicle area corresponding to a common battery installation location is used as the vehicle battery associated area. The adjacent areas of the charging cable at the vehicle end and the adjacent areas of the charging cable at the power supply end are formed according to preset adjacent ranges, based on both ends of the charging cable.

[0037] The vehicle battery associated area can be defined within the vehicle area based on relative position. For example, the area under the seat, under the pedals, the middle of the vehicle body, or a localized area where the identifiable battery compartment cover is located can be designated as the vehicle battery associated area. For different vehicle models, the vehicle attitude can be determined first based on the wheel positions, body contours, and seat positions, and then the vehicle battery associated area can be determined according to the relative coordinates of the vehicle area.

[0038] The source-end anchoring module is used to identify smoke candidate regions and flame candidate regions in consecutive image frames, generate smoke and fire candidate trajectories, determine candidate source ends within the initial appearance window of the smoke and fire candidate trajectory, and determine whether the candidate source ends form a source-end anchoring relationship with the risk anchoring region in the same valid charging link snapshot. The process of determining candidate source ends within the initial appearance window for fireworks candidate trajectories can be found in [reference needed]. Figure 5 Smoke candidate regions can be determined based on gray-white, gray-black, or semi-transparent areas in the image, the degree of edge blurring, local gray-scale changes, and regional diffusion patterns; flame candidate regions can be determined based on red, orange, or yellow highlighted areas, abrupt changes in brightness, irregular contour changes, and continuous frame flickering features. Both smoke and flame candidate regions retain their region boundaries, center positions, areas, identification categories, and the image frame numbers they belong to.

[0039] Smoke and flame candidate regions are only used as input for the generation of subsequent fire and smoke candidate trajectories, and are not directly used as the fire and smoke risk assessment results. Only after the candidate regions form cross-frame fire and smoke candidate trajectories and establish a source-end anchoring relationship with the risk anchoring region in the corresponding valid charging link snapshot will they enter the determination process of valid associated frames or counter-evidence frames.

[0040] Firework candidate trajectories are formed by associating adjacent smoke or flame candidate regions in consecutive image frames that are morphologically continuous or have the same identification category; the initial appearance window consists of several consecutive image frames after the fireworks candidate trajectory is first identified. A source-end anchoring relationship is determined to be formed when the candidate source end and the risk anchoring region in the same valid charging link snapshot satisfy any one of the following: positional overlap, proximity at a preset distance, or stable proximity in consecutive frames.

[0041] When multiple smoke or flame candidate regions exist in the same image frame, candidate regions with consistent identification categories, close center locations, continuous area changes, and corresponding to the same link identifier are preferentially associated as the same smoke / fire candidate trajectory; candidate regions that do not meet the above association conditions are either established as new smoke / fire candidate trajectories or retained as independent candidate regions.

[0042] The initial appearance window can be determined according to a preset number of frames or a preset time length, and its starting point is the image frame in which the candidate fireworks trajectory first appears. The initial appearance window is used to determine the candidate source end of the candidate fireworks trajectory, and subsequent image frames are used to determine the continuous association status between the candidate source end and the risk anchoring area.

[0043] The source-end anchoring module determines the candidate source end based on the boundary of the smoke candidate region or the flame candidate region on the side closest to the risk anchoring region within the initial appearance window, the start position of consecutive frames, or the direction of expansion. When the smoke or flame candidate region has multiple selectable source positions within the initial appearance window, the position that is closer to the risk anchoring region and has less positional change in consecutive image frames is selected as the candidate source. When the candidate region shows an expanding trend, the position in the opposite direction of its expansion is used as the position in the reverse direction of expansion.

[0044] The evidence frame generation module is used to record image frames that form a source-end anchoring relationship as valid association frames, and image frames that do not maintain the source-end anchoring relationship and meet the preset interference characteristics as counter-evidence frames; Both valid association frames and counter-evidence frames are bound to the same fireworks candidate trajectory and the same link identifier. For the same fireworks candidate trajectory, if the candidate source end in the image frame continuously satisfies the source end anchoring relationship, the image frame enters the set of valid association frames; if the candidate source end does not maintain the source end anchoring relationship and simultaneously satisfies the preset interference characteristics, the image frame enters the set of counter-evidence frames.

[0045] The process of generating a set of valid associated frames based on source-end anchoring relationships can be found in [reference needed]. Figure 6 Both valid association frames and counter-evidence frames record the frame number, firework candidate trajectory identifier, link identifier, candidate source location, corresponding risk anchoring area, and frame status marker. The frame status marker for a valid association frame is "source anchoring valid," while the frame status marker for a counter-evidence frame is "source anchoring invalid," and the corresponding preset interference characteristics are recorded.

[0046] Preset interference characteristics include a stable target position without a source-end anchoring relationship, stable target brightness with contour changes below preset change conditions, or a target trajectory continuously deviating from the corresponding valid charging link snapshot.

[0047] The stability of the target position can be determined by the offset of the target center position or the offset of the boundary position in consecutive image frames; the stability of the target brightness can be determined by the change in the brightness of the target area in consecutive image frames; the contour change is lower than the preset change condition can be determined by the change in the target boundary, the change in area, or the degree of contour overlap.

[0048] When the same target satisfies the conditions of stable position, stable brightness, or contour change below the preset change conditions in consecutive image frames, and its candidate source end does not form a source end anchoring relationship with the risk anchoring area, the corresponding image frame will be treated as a counter-evidence frame and will not be used as a positive accumulation basis for the confidence of the firework evolution.

[0049] The risk assessment module is used to accumulate the confidence level of fireworks evolution based on valid associated frames, and to generate a fireworks risk assessment result with a bound link identifier based on the accumulation of confidence level of fireworks evolution attenuated or interrupted by counter-evidence frames.

[0050] The risk assessment module performs forward accumulation of candidate fireworks trajectories in valid associated frames and reverse attenuation of candidate fireworks trajectories in counter-evidence frames. When counter-evidence frames appear consecutively and reach a preset number of frames, the accumulation of fireworks evolution confidence for the corresponding candidate fireworks trajectory is interrupted. When the fireworks evolution confidence reaches the preset risk assessment condition, a fireworks risk assessment result bound to the corresponding link identifier is generated.

[0051] The process of accumulating firework evolution confidence based on effective associated frames can be found in [reference]. Figure 7 Forward accumulation refers to increasing the confidence level of a candidate fireball trajectory when it meets smoke or flame evolution characteristics. Reverse decay refers to decreasing the confidence level of a candidate fireball trajectory when a counter-evidence frame appears. Interruption processing refers to stopping the risk accumulation process of a candidate fireball trajectory when counter-evidence frames appear consecutively for a preset number of frames, or when the candidate source continuously deviates from the corresponding risk anchoring area. The preset risk judgment conditions include at least two of the following: the fireball evolution confidence level reaches a preset risk threshold, the number of valid associated frames reaches a preset number, and the number of counter-evidence frames does not reach a preset exclusion condition.

[0052] The process of attenuating or interrupting the confidence level of the fireworks evolution based on the counter-evidence frames can be found in [link to documentation]. Figure 8 Reverse attenuation can be performed using fixed deduction, proportional deduction, or risk state downgrading. When the confidence level of the fireworks evolution falls below the preset retention condition after reverse attenuation, the candidate fireworks trajectory is transferred to continuous observation. After interruption processing, the subsequent forward accumulation of the candidate fireworks trajectory can be stopped, and the formed valid correlation frames and counter-evidence frames are retained as historical judgment records.

[0053] The risk assessment module accumulates the smoke evolution confidence score for smoke candidate trajectories in the effective associated frames based on area growth, edge diffusion, transparency changes, or center position migration, and accumulates the flame evolution confidence score for flame candidate trajectories in the effective associated frames based on brightness fluctuations, contour jumps, color changes, or area jumps.

[0054] The area growth, edge diffusion, transparency changes, and center position migration of smoke candidate trajectories are determined by the changes in smoke candidate regions between adjacent valid associated frames; the brightness fluctuations, contour jumps, color changes, and area jumps of flame candidate trajectories are determined by the changes in flame candidate regions between adjacent valid associated frames. Image frames that do not enter the valid associated frames do not participate in the positive accumulation of smoke evolution confidence or flame evolution confidence.

[0055] The accumulation of smoke evolution confidence and flame evolution confidence is limited to image frames with the same smoke and fire candidate trajectory, the same link identifier, and a valid source-end anchoring relationship. When the smoke and fire candidate trajectory crosses different link identifiers, or the candidate source end moves to a non-corresponding risk anchoring area, the corresponding trajectory is re-established or processed as a counter-evidence frame.

[0056] The present invention also includes an alarm output module, which outputs alarm information including link identifier, candidate source location, risk anchoring area, valid association frame, counter-evidence frame and fire evolution confidence when generating the fire risk determination result.

[0057] The structure of the smoke and fire risk assessment results and alarm output information can be found in [reference needed]. Figure 9 The alarm information may also include the alarm time, risk trigger frame number, corresponding charging area location, and image frame segments before and after the alarm, which are used to record the image basis in the process of determining the fire risk.

[0058] 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 smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition, characterized in that, It includes an image acquisition module, a link snapshot generation module, a source end anchoring module, an evidence frame generation module, and a risk assessment module; The image acquisition module is used to acquire continuous image frames of the charging area of ​​two-wheeled electric vehicles; The link snapshot generation module is used to identify the vehicle area, charging socket area, charger area and charging cable area in continuous image frames, and generate a valid charging link snapshot with a link identifier when the two ends of the charging cable area are respectively adjacent to the vehicle area, charging socket area or charger area and are stable across frames. The source-end anchoring module is used to identify smoke candidate regions and flame candidate regions in consecutive image frames, generate smoke and fire candidate trajectories, determine candidate source ends within the initial appearance window of the smoke and fire candidate trajectory, and determine whether the candidate source ends form a source-end anchoring relationship with the risk anchoring region in the same valid charging link snapshot. The evidence frame generation module is used to record image frames that form a source-end anchoring relationship as valid association frames, and image frames that do not maintain the source-end anchoring relationship and meet the preset interference characteristics as counter-evidence frames; The risk assessment module is used to accumulate the confidence level of fireworks evolution based on valid associated frames, and to generate a fireworks risk assessment result with a bound link identifier based on the accumulation of confidence level of fireworks evolution attenuated or interrupted by counter-evidence frames.

2. The smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: A valid charging link snapshot includes the same vehicle area, the vehicle end of the charging cable adjacent to the vehicle area, the charging socket area adjacent to the power supply end of the charging cable, and the charger area located between the vehicle area and the charging socket area.

3. The smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: The link snapshot generation module maintains, updates, or releases the link identifier based on the overlap between the vehicle area, charging cable area, charging socket area, and charger area in adjacent image frames, as well as the proximity of the ends of the charging cable.

4. The smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: The risk anchoring area includes at least two of the following: the vehicle battery associated area, the area near the vehicle end of the charging cable, the area near the power supply end of the charging cable, the charger area, and the charging socket area.

5. A smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: The source-end anchoring module determines the candidate source end based on the boundary of the smoke candidate region or the flame candidate region on the side closest to the risk anchoring region within the initial appearance window, the starting position of consecutive frames, or the position in the expansion direction.

6. The smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: When the evidence frame generation module is located within the risk anchoring area or the distance between the candidate source and the risk anchoring area meets the preset proximity condition, the corresponding image frame is recorded as a valid associated frame.

7. The smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: The risk assessment module accumulates the smoke evolution confidence score for smoke candidate trajectories in the effective associated frames based on area growth, edge diffusion, transparency changes, or center position migration, and accumulates the flame evolution confidence score for flame candidate trajectories in the effective associated frames based on brightness fluctuations, contour jumps, color changes, or area jumps.

8. A smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: Preset interference characteristics include a stable target position without a source-end anchoring relationship, stable target brightness with contour changes below preset change conditions, or a target trajectory continuously deviating from the corresponding valid charging link snapshot.

9. A smoke and fire detection system for a two-wheeled electric vehicle charging area based on image recognition according to claim 1, characterized in that: It also includes an alarm output module, which outputs alarm information including link identifier, candidate source location, risk anchoring area, valid association frame, counter-evidence frame and fire evolution confidence level when generating fire risk assessment results.

Images