A method and system for offline navigation of intelligent terminals in dense forest environments

Abstract

This application proposes an offline navigation method and system for smart terminals in dense forest environments. The method includes: performing multimodal scanning of the surrounding dense forest environment to obtain first collected data, and filtering candidate features from the first collected data in conjunction with environmental sensor data; determining the target guidance area and collection mode based on the candidate features, and adjusting the collection parameters of the data collection module according to the collection mode; guiding multimodal scanning to obtain second collected data, verifying the candidate features based on the second collected data, and obtaining effective features; deriving the relative orientation based on the effective features, and outputting the navigation result in conjunction with the local offline map. This application relies on the hardware characteristics of smart terminals to construct a dedicated offline navigation logic for dense forests. By mining the natural features commonly found in dense forest environments as the core basis for navigation, and combining multimodal collection with environmental adaptation optimization, it achieves high-precision, high-stability, and long-endurance offline navigation in dense forest environments.

Description

Technical Field

[0001] This application relates to the field of outdoor navigation, and more specifically, to an offline navigation method and system for smart terminals in dense forest environments. Background Technology

[0002] With the increasing prevalence of outdoor adventures, forestry surveys, and mountain search and rescue activities, the demand for accurate navigation in dense forest environments is becoming increasingly urgent. Dense forest areas typically feature dense canopies, complex terrain, and severe obstruction, posing significant challenges to the adaptability of traditional navigation technologies. Existing navigation methods struggle to achieve stable and reliable offline navigation in such scenarios, with specific drawbacks as follows: Current mainstream navigation technologies are based on satellite navigation, which relies on receiving radio signals transmitted by satellites to achieve positioning and navigation. However, in dense forest environments, the thick canopy severely blocks, reflects, and attenuates satellite signals, causing a sharp decrease in signal strength or even complete loss, making positioning impossible. At the same time, the complex terrain and tall vegetation in dense forests can also cause multipath effects. Even if some signals can be received, signal interference will cause a significant decrease in positioning accuracy, with errors reaching several meters to tens of meters, which completely fails to meet navigation requirements.

[0003] To compensate for the shortcomings of satellite navigation, inertial navigation and magnetic field navigation are existing supplementary solutions. Inertial navigation relies on sensors such as accelerometers and gyroscopes, calculating position and direction through integration calculations, without requiring external signal support. However, this solution has inherent drawbacks: its positioning error accumulates over time, and the error increases dramatically after prolonged continuous use. In scenarios requiring navigation deep into dense forests for extended periods, it cannot maintain effective navigation accuracy. Furthermore, inertial navigation has high hardware precision requirements, and the sensors on ordinary mobile devices are insufficient to meet the long-term navigation needs in dense forest environments. Magnetic field navigation senses direction based on the Earth's magnetic field, but it is highly susceptible to interference from the dense forest environment. Magnetic rocks, soil composition, and metal detection targets in dense forests can distort the local magnetic field, causing data collected by the magnetic field sensor to be distorted, making it impossible to accurately determine north-south direction, and further compromising navigation capabilities.

[0004] Therefore, there is an urgent need for a navigation method that can fully explore the characteristics of the natural environment of dense forests to meet the navigation needs of various outdoor activities in dense forest scenarios. Summary of the Invention

[0005] To address the problems existing in current technologies, this application provides an offline navigation method and system for smart terminals in dense forest environments. The specific solution is as follows: An offline navigation method for smart terminals in a dense forest environment includes: The data acquisition module of the smart terminal performs multimodal scanning of the surrounding dense forest environment to obtain the first collection data, and combines the synchronously acquired environmental sensor data to filter out candidate features related to geographical location from the first collection data. Based on the candidate features, the target guidance area that needs to be focused on and the corresponding acquisition mode are determined. A visual acquisition guide is generated based on the target guidance area, and the acquisition parameters of the data acquisition module are adjusted according to the acquisition mode. The user is guided to perform multimodal scanning of the target guidance area according to the visual acquisition guide to obtain the second acquisition data. The candidate features are verified based on the second acquisition data, and the verified candidate features are taken as effective features that are strongly correlated with geographic location. Based on the aforementioned effective features, the relative orientation is derived, and the navigation result is output in conjunction with the local offline map.

[0006] In some specific embodiments, the first acquired data includes visual image data and spectral data; the environmental sensing data includes temperature and humidity, air pressure, and ambient light intensity data, and is associated with the first acquired data by timestamp. Based on visual image data and spectral data, potential feature regions are divided, including at least plant areas and light and shadow areas. From each potential feature region, features related to geographical orientation are extracted to obtain candidate features.

[0007] In some specific embodiments, for plant areas, spectral data is used to capture the reflectance characteristics of leaf chlorophyll-related bands, and leaf outline and orientation information are extracted by combining visual images. The plant activity is determined with reference to the temperature and humidity data. When the activity is determined to be high, the priority of this type of plant feature in the candidate feature is increased; when the activity is determined to be low, the priority of this type of plant feature in the candidate feature is decreased. For the light and shadow area, combined with ambient light intensity and air pressure data, regular-shaped light spots and blurry short shadows are excluded, and only irregular sun spots and clear shadows of fixed objects are retained. It is also ensured that the light and shadow distribution initially matches the solar azimuth interval determined based on the local clock to obtain the light and shadow features, and these light and shadow features are included in the candidate features.

[0008] In some specific embodiments, determining the target guidance region based on candidate features specifically includes: Locate the core distribution area of ​​candidate features in the dense forest environment, and delineate the target guidance area based on feature clarity and area ratio to ensure that the target guidance area covers the core part of the candidate features.

[0009] In some specific embodiments, if the candidate features involve plant features, the acquisition mode is switched to spectral focusing mode; In the spectral focusing mode: the hyperspectral acquisition unit is adjusted to focus on the chlorophyll-related band, and interference from other bands is shielded; the camera exposure parameters are optimized, the focus range is fixed to the distance corresponding to the target guidance area, and the white balance and sharpness are adjusted to enhance the leaf outline and texture details, highlighting the difference between the light-facing and shaded sides.

[0010] In some specific embodiments, if the candidate features involve light and shadow features, the acquisition mode is switched to the light and shadow enhancement mode; In the light and shadow enhancement mode: control the camera exposure, switch to spot metering mode to focus on the core area of ​​light and shadow, increase contrast to widen the grayscale difference between the light spot and the background, and between the shadow and the object, and fix the shutter speed to eliminate the light and shadow motion caused by the leaf shaking.

[0011] In some specific embodiments, when the light and shadow features and plant features are insufficient, a potential feature region, namely a rock region, is delineated based on visual image data. Rock features are obtained by extracting the direction information of natural weathering textures on the rock surface, and the rock features are included as supplementary candidate features. If the candidate features involve rock features, switch the acquisition mode to texture enhancement mode; In the texture enhancement mode: the white balance is adjusted to weaken the color interference of soil and decaying leaves, the sharpness and contrast are improved to highlight the direction and depth of weathering texture, and the focus range is locked to the core area of ​​the rock to ensure clear acquisition of texture details.

[0012] In some specific embodiments, verifying the candidate features based on the second collected data includes: If the candidate feature is a plant feature, the difference in spectral reflectance between the light-facing and shaded sides of the leaves in the second collection data is analyzed. At the same time, the distribution ratio of the leaves in the flat and tilted states is counted. When the reflectance difference has stable distinguishability, or the distribution of leaf orientation shows a consistent trend, and this trend matches the solar orientation pattern, the candidate plant feature is determined to pass the orientation correlation verification. If the candidate feature is a light and shadow feature, then, in conjunction with the solar azimuth interval determined by the local clock, it is verified whether the regional distribution difference of light spot density in the second collection data corresponds to the solar azimuth. At the same time, it is verified whether the extension direction of the shadow of the fixed object is inversely related to the solar azimuth. If at least one of the two verifications satisfies the rule, the candidate feature of light and shadow is determined to have passed the azimuth correlation verification.

[0013] In some specific embodiments, verifying the candidate features based on the second collected data further includes: The stability of candidate features against environmental interference was verified from both temporal and spatial dimensions. Extract multiple consecutive frames of data from the second collection data, calculate the fluctuation range of the core indicators of the candidate features, and if the fluctuation range is within a preset reasonable range and there is no drastic change caused by short-term interference, it is determined to pass the time dimension verification. The target guidance area is divided into multiple equal sub-regions, and the core indicators of candidate features are extracted from each sub-region. If the difference between the indicators of each sub-region is within the preset threshold and there is no deviation of the indicators caused by local abnormal interference points, it is determined to pass the spatial dimension verification.

[0014] An offline navigation system for smart terminals in a dense forest environment, comprising: The first acquisition unit is used to perform multimodal scanning of the surrounding dense forest environment through the data acquisition module of the smart terminal to obtain the first acquisition data, and to filter out candidate features related to geographical location from the first acquisition data in combination with the synchronously acquired environmental sensing data. The parsing unit is used to determine the target guidance area that needs to be focused on and the corresponding acquisition mode based on the candidate features, generate a visual acquisition guide based on the target guidance area, and adjust the acquisition parameters of the data acquisition module according to the acquisition mode. The second acquisition unit is used to guide the user to perform multimodal scanning of the target guidance area according to the visual acquisition guide to obtain the second acquisition data, verify the candidate features based on the second acquisition data, and take the verified candidate features as effective features that are strongly correlated with geographical location. The fusion unit is used to deduce the relative orientation based on the effective features and output navigation results by combining them with the local offline map.

[0015] Beneficial effects: This application proposes an offline navigation method and system for smart terminals in dense forest environments. It builds a dedicated offline navigation logic for dense forests based on the hardware characteristics of smart terminals, without relying on external satellite signals and network support. It uses the natural features commonly found in dense forest environments as the core basis for navigation, and combines multimodal acquisition and environmental adaptation optimization to achieve high-precision, high-stability, and long-endurance offline navigation in dense forest environments. It effectively solves problems such as the loss of satellite navigation signals, accumulation of inertial navigation errors, and distortion caused by interference in magnetic field navigation. Even deep within dense forests with dense canopies, magnetic rock interference, and complex, undulating terrain, the system can continuously provide reliable navigation guidance, significantly improving the safety and efficiency of navigation in dense forest environments for outdoor adventures, forestry surveys, and mountain search and rescue activities. Simultaneously, the solution uses multimodal scanning combined with environmental sensor data to filter candidate features, adapting to dynamic changes in temperature, humidity, light, and air pressure within the dense forest environment. This allows for accurate identification of directional features while avoiding invalid interfering features, enhancing the targeting and reliability of feature extraction. This addresses the shortcomings of existing solutions that cannot dynamically adjust to real-time environmental conditions and suffer from blind feature recognition. Based on candidate features, the system determines the target guidance area and adapts to a dedicated acquisition mode and adjusts acquisition parameters, enabling precise acquisition of core feature areas. This strengthens the distinction between target features and background, effectively overcoming the challenges posed by decaying leaves and other debris within dense forests. Interference from objects and vegetation obstructions is mitigated, improving the clarity and effectiveness of feature collection. Compared to general collection modes, this significantly reduces the proportion of invalid data and decreases computational power consumption. Candidate features are validated using secondary collection data, ensuring their effectiveness through both location correlation and anti-interference stability. False features caused by short-term gusts, miasma obstruction, and local anomalies are eliminated, ensuring the reliability of navigation data and avoiding navigation deviations caused by misjudgments of single features. Finally, relative orientation is derived based on effective features and combined with local offline maps to output navigation results. This approach achieves accurate direction determination based on natural features and avoids impassable areas such as steep slopes and rivers by combining offline terrain data, providing dual protection for direction guidance and path optimization. It is compatible with the low-power outdoor requirements of smart terminals, requiring no additional dedicated hardware, thus balancing navigation performance and device battery life.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the offline navigation method for smart terminals according to this application; Figure 2 This is a schematic diagram illustrating the principle of the offline navigation method for smart terminals in this application; Figure 3 This is a schematic diagram of the candidate feature selection process in this application; Figure 4 This is a schematic diagram of the data acquisition parameter adjustment process in this application; Figure 5 This is a schematic diagram of the offline navigation system module for the smart terminal of this application.

[0019] Reference numerals in the attached figures: 1-First acquisition unit; 2-Analysis unit; 3-Second acquisition unit; 4-Fusion unit. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] This application proposes an offline navigation method for smart terminals in dense forest environments. It deeply integrates projection technology with outdoor navigation, breaking the limitations of traditional outdoor navigation that relies on handheld devices or wearable screens. Navigation information is directly presented on the ground in front of the user in projection form, constructing a complete collaborative link. This fully leverages the intuitive and convenient information output advantages of projection technology while precisely matching the dynamic and complex environmental requirements of outdoor navigation. A flowchart of the adaptive projection navigation method is attached. Figure 1 As shown in the attached diagram, the principle is as follows: Figure 2 As shown, the specific solution is as follows: An offline navigation method for smart terminals in a dense forest environment includes: 101. The data acquisition module of the smart terminal performs multimodal scanning of the surrounding dense forest environment to obtain the first collection data, and combines the synchronously acquired environmental sensing data to filter out candidate features related to geographical location from the first collection data; 102. Based on candidate features, determine the target guidance area that needs to be focused on and the corresponding acquisition mode. Generate a visual acquisition guide based on the target guidance area and adjust the acquisition parameters of the data acquisition module according to the acquisition mode. 103. Guide users to perform multimodal scanning of the target guidance area according to the visual collection guide to obtain the second collection data. Verify candidate features based on the second collection data, and take the verified candidate features as effective features that are strongly correlated with geographic location. 104. Derive relative orientation based on effective features and output navigation results by combining local offline maps.

[0022] In this application, the smart terminal is exemplified by a rugged phone. A rugged phone is a special type of mobile terminal with three core protective capabilities: waterproof, dustproof, and shockproof. It is designed for harsh environments such as outdoor adventure, forestry survey, mountain search and rescue, and industrial operations. Its core feature is that through hardware structure optimization and protection technology upgrades, it can resist damage to the device from complex environments, while retaining the core functions of a regular mobile phone, such as communication, data acquisition, and computing. It is a dedicated smart device adapted to extreme scenarios such as dense forests, deserts, and mountains.

[0023] The proposed solution requires the following basic preparatory work to ensure smooth progress at each stage: First, satellite signal and navigation mode prediction: the smart terminal automatically detects satellite signal strength. If the signal is insufficient due to dense forest obstruction, making positioning calculation impossible, or if the signal is frequently interrupted or severely interfered with, this offline navigation method is triggered. If the satellite signal is stable, traditional satellite navigation is used first, achieving adaptive mode switching. Second, local offline map preparation: users need to download offline map data of the target dense forest area and its surroundings in advance. The map must include core information such as terrain undulations, rivers, steep slopes, and paths to ensure that map data can be used to assist navigation in offline mode. At the same time, the terminal automatically updates the map cache and clears redundant data to ensure smooth map loading. Third, hardware module self-... The system performs the following steps: First, it checks and activates the smart terminal, automatically starting the data acquisition module, environmental sensing module, processor, and visualization output module (such as a projector or screen). It checks whether each module is working properly, ensuring that there are no faults in data acquisition, signal transmission, processing, and guidance output. At the same time, it calibrates the sensing module to avoid initial errors. Second, it optimizes power consumption and storage, automatically adjusting the hardware power consumption mode, closing unnecessary background programs, and reserving sufficient storage space to cache acquired data, map data, and intermediate processing results. This ensures that the system will not be interrupted due to excessive power consumption or insufficient storage during long-term navigation in dense forests. Third, it calibrates the clock, pre-calibrating the smart terminal's local clock to ensure the accuracy of subsequent judgments on the relationship between natural features and the sun's position, eliminating the need for network clock synchronization.

[0024] Step 101's core is to complete preliminary data collection and feature selection, providing foundational data support for subsequent navigation. The smart terminal's data collection module includes a camera and a spectral acquisition unit. Multimodal scanning simultaneously collects various types of data, such as visual image data and spectral data, covering various scene information in the dense forest environment, including vegetation, terrain, and light and shadow. The first collected data is the raw comprehensive data obtained after the full-domain scan. Simultaneously acquired environmental sensor data includes temperature, humidity, air pressure, and ambient light intensity. Its role is to assist in feature selection, because temperature, humidity, light, and air pressure in the dense forest environment affect the stability and directional orientation of natural features. Combining these data can eliminate features that are greatly affected by environmental interference and have no directional value. Candidate features that may be related to geographical location are selected from the first collected data. These candidate features are natural features that are common in the dense forest environment and have inherent directional potential. It is not random selection; the core is to explore the potential correlation between natural features and geographical location, getting rid of dependence on external signals. At the same time, environmental data assists in the selection, improving the relevance of candidate features and reducing the amount of subsequent computation.

[0025] Step 102 focuses on achieving precise data acquisition adaptation, addressing the issues of ambiguous targets and data redundancy in dense forest environments. Based on the candidate features selected in Step 101, the terminal first locates their core distribution area within the dense forest environment, avoiding interference areas such as decaying leaves, debris, and temporary obstructions. A target guidance area is defined to ensure that this area completely covers the core portion of the candidate features, and that the feature clarity and area proportion meet subsequent acquisition requirements, avoiding an excessively large acquisition range that leads to increased invalid data. The corresponding acquisition mode is a dedicated acquisition logic determined based on the attributes of the candidate features. Its core objective is to enhance the distinction between candidate features and the background environment, improving the signal-to-noise ratio of the feature data. Subsequently, the terminal adjusts the acquisition parameters of the data acquisition module according to this acquisition mode, achieving hardware-level parameter adaptation. Simultaneously, a visual acquisition guide is generated based on the target guidance area. Through projection, screen display, or voice prompts on the smart terminal, the user is clearly informed of the acquisition direction, distance, and area to be focused, guiding the user to accurately align with the target guidance area for acquisition. This solves the problem of obstructed vision and difficulty in accurately locating the acquisition target in dense forest environments, laying the foundation for obtaining high-quality data subsequently.

[0026] Step 103's core is to accurately verify candidate features, eliminate interfering features, and ensure the reliability of navigation data. Guiding users to operate according to the visual data collection guide essentially involves having users perform targeted multimodal scanning of the target guidance area under terminal guidance. Compared to the full-area scanning in Step 101, this scan is more focused and precise, resulting in high-quality data around the core areas of the candidate features, effectively reducing irrelevant interference data. Verifying candidate features based on this second-collected data involves determining whether the candidate features truly have a strong correlation with geographical location, rather than being false features formed by temporary interference factors in the dense forest environment. Verification eliminates invalid features caused by gusts of wind, miasma obstruction, or local debris interference, retaining only features with stable directional orientation as valid features. This step is crucial for ensuring the accuracy of subsequent direction derivation, avoiding navigation errors due to deviations in candidate features.

[0027] Step 104 is crucial for directional derivation and navigation result output, transforming feature data into practical navigation guidance. Deriving relative orientation based on effective features utilizes the inherent correlation between effective features and geographical location. The intelligent terminal processor analyzes the effective feature data to determine core relative orientation information such as north-south direction, without relying on magnetic field sensors or satellite signals, effectively avoiding magnetic field interference and signal blockage issues in dense forest environments. Combining the navigation results with local offline maps integrates the derived relative orientation with terrain information from the offline map. This not only provides directional guidance but also allows users to avoid impassable areas such as rivers and steep slopes based on the map's terrain data, planning feasible routes. Furthermore, the navigation results are displayed visually (e.g., arrow indicators, path markings), ensuring users can clearly obtain guidance in dense forest environments, achieving accurate and safe offline navigation.

[0028] In some specific embodiments, the first acquired data includes visual image data and spectral data; the environmental sensing data includes temperature and humidity, air pressure, and ambient light intensity data, and is associated with the first acquired data by timestamp; based on the visual image data and spectral data, at least potential feature regions including plant areas and light and shadow areas are divided, and candidate features are obtained by extracting features related to geographical orientation from each potential feature region.

[0029] The visual image data in the first data acquisition was obtained by the rugged phone's camera module through multi-frame continuous shooting of the surrounding dense forest environment. This data can intuitively present scene details such as vegetation, terrain, and light and shadow, providing a morphological basis for subsequent area division. The spectral data was obtained by scanning with the hyperspectral acquisition unit on the rugged phone. This data can capture the reflection and absorption characteristics of different substances to different wavelengths of light. In particular, it can accurately identify the spectral signals of the chlorophyll-related wavelengths in leaves. This is the key to distinguishing plants from other scene elements and capturing the differences between plants in light and shade, making up for the inability to perceive the intrinsic properties of substances by relying solely on visual images.

[0030] The environmental sensing data, including temperature, humidity, air pressure, and ambient light intensity, are collected in real-time by the phone's built-in temperature and humidity sensor, air pressure sensor, and light sensor, respectively. These data are linked to the first collected data by timestamp, essentially establishing a correspondence between the environmental state at the time of collection and the corresponding image and spectral data. This ensures that subsequent feature analysis can be performed in conjunction with the environmental conditions at that time, avoiding misjudging feature value out of context. Potential feature regions are segmented based on visual image data and spectral data. The core principle is to utilize the complementarity of these two types of data to achieve precise partitioning. Through image segmentation algorithms combined with spectral feature thresholds, plant areas and light and shadow areas are separated from complex dense forest scenes. Plant areas are identified based on visual leaf morphology and outline, as well as chlorophyll characteristic band signals in the spectrum. Light and shadow areas are defined based on visual differences in brightness contrast and spectral brightness band signals. These two types of areas are the most common in dense forest environments and are naturally associated with geographical orientation. Selecting them as potential feature regions ensures the universality and specificity of candidate features.

[0031] Extracting candidate features related to geographic orientation from each potential feature region is the core foundation for subsequent navigation. For plant areas, the contour distribution, orientation angle, and reflectance differences of chlorophyll-related bands of leaves are extracted. These features exhibit regular distribution due to the phototropism of plants, indirectly indicating the sun's position and thus determining geographic orientation. For light and shadow areas, the distribution range, density, and extension direction and length of light spots are extracted. These features are influenced by the sun's position and have clear spatiotemporal patterns, directly relating to geographic orientation. The design of timestamp association is crucial, ensuring that when analyzing visual images and spectral data at a certain moment, the corresponding temperature, humidity, air pressure, and ambient light data can be simultaneously accessed. For example, temperature and humidity can be used to determine plant activity; plants with high activity show more significant spectral differences between light and shadow, and their features have higher priority. Ambient light intensity and air pressure are used to determine the effectiveness of light and shadow, avoiding misjudging abnormal light and shadow in extreme environments such as rain or strong light as effective features. Ultimately, this allows for the selection of candidate features with navigation potential from complex dense forest scenes, providing a reliable initial basis for subsequent accurate navigation.

[0032] In some specific embodiments, for plant areas, spectral data is used to capture the reflectance characteristics of chlorophyll-related bands in leaves, and leaf outline and orientation information are extracted by combining visual images. Temperature and humidity data are referenced to determine plant activity. When activity is determined to be high, the priority of this type of plant feature in the candidate feature list is increased; when activity is determined to be low, the priority of this type of plant feature in the candidate feature list is decreased. The candidate feature screening process is attached. Figure 3 As shown.

[0033] The reflection and absorption patterns of chlorophyll on specific wavelengths of light are species-wide, and the phototropism of plants leads to differences in chlorophyll activity between the light-facing and shaded sides, resulting in a clear distinction in spectral reflectance signals. This is the core basis for associating plant characteristics with geographical orientation. By using a rugged mobile phone's hyperspectral acquisition unit to lock onto this wavelength, the intrinsic optical differences of leaves can be accurately captured from complex vegetation backgrounds, overcoming the limitation of visual images that can only identify external morphology. Combining visual images to extract leaf contour and orientation information involves image segmentation and morphological recognition algorithms to delineate leaf boundaries from the visual image, determine the growth orientation of individual leaves and leaf clusters, and combine intrinsic differences at the spectral level with external orientation at the morphological level, making plant characteristics more directional. Judging plant activity by referring to temperature and humidity data is essentially based on plant growth characteristics. Under suitable temperature and humidity conditions, plants have vigorous metabolism and high chlorophyll activity, and the difference in spectral reflectance between the light-facing and shaded sides is more stable and significant. Such characteristics are highly reliable as a navigation basis, so their priority in being included as candidate features is increased. On the other hand, under extreme temperature and humidity conditions, plant activity decreases, the difference in chlorophyll reflectance weakens or even becomes abnormal, and the feature's directionality decreases significantly. Therefore, its priority is reduced. Through this dynamic adjustment, plant characteristics that are more suitable for the current environment can be screened out, avoiding invalid features from interfering with subsequent processes.

[0034] For the light and shadow area, based on ambient light intensity and air pressure data, regularly shaped light spots and blurry short shadows are excluded, retaining only irregular solar spots and clear shadows of fixed objects. It is also ensured that the light and shadow distribution initially matches the solar azimuth interval determined based on the local clock, thus obtaining the light and shadow features, which are then included in the candidate features. The candidate feature selection process is attached. Figure 3 As shown.

[0035] In dense forest environments, regularly shaped light spots are mostly reflections from non-natural objects such as metallic debris and smooth rocks, offering no value for geographical orientation. Blurry, short shadows are often formed by branches swaying in the breeze, exhibiting extremely poor stability and unusable as effective navigational information. Ambient light intensity data can help determine the effectiveness of light; in strong light, the boundaries between sunlight spots and shadows are clear and stable, while in weak light or backlight, light and shadow are easily distorted. Barometric pressure data can indirectly predict weather conditions; on cloudy or rainy days, light and shadow characteristics are blurred and irregular. Based on these two types of data, the aforementioned interfering light and shadow can be accurately eliminated. Only irregular sunlight spots and clear shadows of fixed objects are retained because sunlight-formed light spots are mostly irregular in shape, while the shadows of fixed objects (such as tall trees and rocks) are controlled by the sun's position, exhibiting clear boundaries, stable shapes, and well-defined spatiotemporal patterns, directly correlated with geographical orientation. Ensuring a preliminary match between the light and shadow distribution and the solar azimuth range determined by the local clock involves utilizing the fixed pattern of solar azimuth changes over time. The current time is obtained through the local clock of a rugged phone, and combined with the preset solar azimuth operation pattern, it is determined whether the light and shadow distribution (such as the direction of light spot density concentration and the direction of shadow extension) matches the solar azimuth at that moment. For example, if the sun is in the east in the morning, the shadow should extend to the west. Through this preliminary matching, abnormal light and shadow are further eliminated, ensuring that the light and shadow features included in the candidate features have reliable geographical orientation correlation attributes, providing a high-quality foundation for subsequent feature verification and azimuth deduction.

[0036] In some specific embodiments, determining the target guidance area based on candidate features specifically includes: locating the core distribution area of ​​candidate features in a dense forest environment, and delineating the target guidance area by combining feature clarity and area proportion, ensuring that the target guidance area covers the core part of the candidate features. Locating the core distribution area of ​​candidate features in a dense forest environment essentially means identifying the area where candidate features (plant features, light and shadow features) are most concentrated, have the strongest signal, and are most directional from the complex scene of the first data collection. For plant features, the core distribution area is a vegetation area with dense leaves, significant differences in chlorophyll spectral reflectance, and a regular and concentrated leaf orientation, rather than the edge area with scattered branches and leaves; for light and shadow features, the core distribution area is an area with concentrated light spot density, clear shadow boundaries, and no obvious occlusion. The rugged phone's data processing unit performs coordinate positioning and density analysis on the candidate features, marking the area with the highest feature concentration and the most complete information, providing a basis for the delineation of the guidance area. Defining the guidance area based on three principles—feature clarity, area proportion, and non-interference—is crucial to ensuring its effectiveness. The feature clarity principle requires eliminating blurred feature areas caused by overlapping branches and leaves, decaying leaves, and light refraction, retaining only those with complete outlines, stable spectral signals, and sharp light and shadow boundaries. Image clarity assessment algorithms and spectral signal intensity detection ensure accurate extraction of feature information within the guidance area. The area proportion principle balances feature coverage and acquisition efficiency. The guidance area must be large enough to cover the complete distribution of core features, avoiding the loss of core features due to an excessively small area. Simultaneously, the area size should be controlled to avoid including too many irrelevant areas that cause data redundancy, and should be within a reasonable range suitable for the field of view of rugged mobile phones. The non-interference principle avoids interference sources in dense forest environments, including piles of decaying leaves, man-made debris, and areas of swaying branches and leaves. Feature recognition algorithms distinguish candidate features from interfering elements, ensuring that the guidance area contains only the target candidate features, without additional interfering elements affecting secondary acquisition. Ultimately, ensuring the targeted guidance area covers the core part of the candidate features is essential for enabling secondary acquisition to accurately capture the most valuable feature information. Whether it's the core of the spectral difference between light and shadow in plant characteristics, the concentrated area of ​​leaf orientation, or the core of the regular distribution of light and shadow characteristics, all can be fully included in the collection field of view, providing high-quality and highly targeted data support for subsequent feature verification based on the second collection data, reducing invalid data calculations, improving the efficiency and reliability of the overall navigation process, and adapting to the actual needs of precise collection in dense forest environments.

[0037] In some specific embodiments, if plant features are involved among the candidate features, the acquisition mode is switched to spectral focusing mode. Through targeted hardware parameter optimization, key information related to geographical orientation in plant features is enhanced, background interference from dense forest environments is reduced, and the quality of the second acquisition data is improved, providing an accurate and reliable basis for subsequent feature verification. The acquisition parameter adjustment process is as follows: Figure 4 As shown.

[0038] In spectral focusing mode: adjust the hyperspectral acquisition unit to focus on the chlorophyll-related band and shield other band interference; optimize camera exposure parameters, fix the focus range to the distance corresponding to the target guidance area, adjust white balance and sharpness to enhance leaf outline and texture details, and highlight the difference between the light-facing and shaded sides.

[0039] This mode is designed entirely around the core attributes of plant characteristics. It is a dedicated optimization strategy adapted to the needs of plant feature acquisition, unlike the indiscriminate data acquisition of general acquisition modes. It can significantly improve the acquisition efficiency and accuracy of target features. In the spectral focusing mode, the hyperspectral acquisition unit is adjusted to focus on the chlorophyll-related band and block interference from other bands. The core principle is that chlorophyll has fixed and unique reflection and absorption characteristics for light in specific bands. The spectral signals of these bands are the key basis for distinguishing the light-facing and shaded sides of plants. In dense forest environments, there are various background elements such as soil, decaying leaves, and rocks. The spectral signals of these elements can interfere with plant features. Focusing on the chlorophyll-related band and blocking other bands can filter out spectral noise from irrelevant backgrounds and retain only the effective spectral data directly related to plant features, thereby accurately capturing the spectral reflectance differences between the light-facing and shaded sides of leaves. Optimizing camera exposure parameters is crucial because the lighting in dense forests is complex and unevenly distributed. Overexposure leads to the loss of detail on the illuminated side of the leaves, while underexposure fails to capture the texture features of the shaded side. Adjusting exposure parameters balances the overall brightness of the leaves, ensuring a clear distinction between the illuminated and shaded sides. Fixing the focus range to the corresponding distance of the target guidance area prevents focus drift caused by user-held rugged phones shaking or obstacles in the dense forest, ensuring the captured field of view remains locked on the core distribution area of ​​plant features and avoiding the inclusion of irrelevant data due to focus deviation. Adjusting white balance and sharpness enhances leaf outlines and texture details because the light in dense forests tends to be greenish or dark, causing color distortion in the leaves and affecting outline recognition. Adjusting white balance restores the true color of the leaves, improving the distinction between the outline and the background. Increasing sharpness enhances the edge texture of the leaves, making the boundary between the illuminated and shaded sides clearer and further highlighting their differences. All the above parameter adjustments are not independent, but form a collaborative optimization acquisition strategy. The final effect is that the plant characteristics in the second acquisition data, such as chlorophyll spectrum differences, leaf outlines and orientation information, are significantly enhanced, and irrelevant background interference is reduced to the maximum extent. This provides high-quality and highly directional data support for subsequent verification work such as analyzing the difference in reflection of leaves on the light and shaded sides and statistically analyzing the distribution ratio of leaf orientation. It effectively solves the problems of blurred plant characteristics, severe interference and inability to extract effective orientation information under the general acquisition mode.

[0040] In some specific embodiments, if the candidate features involve light and shadow features, the acquisition mode is switched to the light and shadow enhancement mode; by optimizing the camera acquisition parameters in a targeted manner, the boundary clarity and morphological stability of light and shadow features are enhanced, the damage to light and shadow information caused by complex light and dynamic interference in the dense forest environment is reduced, and the second acquisition data can accurately capture the key features related to the relationship between light and shadow and the sun's position, providing a high-quality basis for subsequent feature verification.

[0041] In the Light and Shadow Enhancement mode: control the camera exposure, switch to spot metering mode to focus on the core area of ​​light and shadow, increase contrast to widen the grayscale difference between the light spot and the background, and between the shadow and the object, and fix the shutter speed to eliminate the light and shadow motion caused by the leaf movement.

[0042] This mode differs from the indiscriminate data acquisition of general acquisition modes, and is designed entirely around the core needs of light and shadow features. The value of light and shadow features lies in their distribution patterns and morphological characteristics controlled by the sun's position. However, in dense forest environments, the light is chaotic and the branches and leaves sway frequently, which can easily lead to blurred light and shadow and distorted shapes. This mode specifically addresses these pain points through parameter co-optimization. In the light and shadow enhancement mode, controlling the camera's exposure is a fundamental prerequisite. The light distribution in dense forests is extremely uneven, with both strong light spots projected through gaps in the tree canopy and dark areas formed by dense vegetation. Overexposure will cause light spot details to overflow and the boundaries to disappear, while underexposure will cause the texture and boundary features inside the shadows to be lost. By precisely controlling the exposure, a balanced adaptation of the brightness of light spots and shadow areas can be achieved, preserving the core shape of the light spots without losing the boundary details of the shadows. Switching to spot metering mode focuses on the core area of ​​light and shadow. The key is to improve metering accuracy and avoid interference from ambient light from surrounding vegetation. Spot metering can precisely lock onto the core area of ​​light and shadow (such as the center of a light spot or the edge of a shadow) for metering, ensuring accurate exposure in that area and clearly presenting the core features of light and shadow, rather than having the brightness flattened by ambient light, thus weakening the features of light and shadow. Increasing contrast to widen the grayscale difference between the light spot and the background, and between shadows and objects, is because the grayscale values ​​of elements such as vegetation and soil in a dense forest background easily overlap with the light and shadow areas, causing blurring of light and shadow boundaries. By increasing contrast, the distinction between the light and shadow areas and the surrounding background can be strengthened, making the light spots more prominent and the shadow boundaries sharper. This facilitates the subsequent extraction of key features such as the distribution range and extension direction of light and shadow, while reducing the interference of background elements on the features of light and shadow. Fixed shutter speed eliminates light and shadow trails caused by leaf movement, optimizing for dynamic interference in dense forest environments. In dense forests, branches and leaves are easily affected by breezes, causing them to sway. If the shutter speed is too slow, the captured light and shadow will exhibit trailing and distorted shapes, completely losing their regular characteristics corresponding to the sun's position. A fixed high shutter speed quickly freezes the light and shadow shapes, eliminating the dynamic interference of swaying branches and leaves, ensuring that the captured light spots and shadows are realistic, stable, and conform to the inherent laws controlled by the sun's position. The above parameter adjustments form a collaborative optimization acquisition strategy, not an isolated operation. The final effect is a significant improvement in the clarity of the boundaries, the stability of the shapes, and the distinction from the background of the light and shadow features in the second acquisition data. This effectively filters out invalid information caused by cluttered light and dynamic interference, accurately preserving the core features associated with the sun's position. This provides highly reliable data support for subsequent verification of the correspondence between light spot distribution and the sun's position, and the inverse correlation between the shadow extension direction and the sun's position, solving the problems of blurred and distorted light and shadow features and the inability to extract effective azimuth information in general acquisition modes.

[0043] In some specific embodiments, when light and shadow features and plant features are insufficient, a potential feature region, the rock region, is delineated based on visual image data. Rock features are obtained by extracting the direction information of natural weathering textures on the rock surface and are included as supplementary candidate features. If the candidate features involve rock features, switch the acquisition mode to texture enhancement mode. In texture enhancement mode: adjust the white balance to weaken the color interference of soil and decaying leaves, improve sharpness and contrast to highlight the direction and depth of weathering texture, lock the focus range to the core area of ​​the rock, and ensure clear acquisition of texture details.

[0044] By constructing a three-tiered candidate feature system encompassing plants, light and shadow, and rocks, this system addresses navigation gaps caused by insufficient or ineffective plant and light and shadow features in dense forest environments. Supplementing with stable rock features ensures navigation continuity, while a dedicated texture enhancement acquisition mode guarantees the effectiveness of the supplementary features. In scenarios where light and shadow and plant features are insufficient, typically in dense forests where thick canopies completely block light, in sparsely vegetated areas with exposed rocks, or in extreme weather conditions that cause a sudden drop in plant activity and complete disappearance of light and shadow, navigation cannot be achieved relying solely on the first two types of features. Therefore, it is necessary to utilize the more stable natural feature of rocks as a supplement.

[0045] Delineating rock regions based on visual image data involves using image segmentation algorithms to identify inherent morphological features of rocks, such as rough texture without leaf-like patterns, uniform color that clearly distinguishes it from soil and decaying leaves, and contiguous areas without vegetation cover. This allows for the precise delineation of potential rock feature areas from complex, dense forest backgrounds. Rock features are obtained by extracting the orientation information of natural weathering textures on the rock surface. The core principle is that rock weathering textures are influenced by regional geological structures and long-term wind and water erosion, resulting in significant regional stability patterns. The orientation of weathering textures within the same region is generally consistent, and this orientation has a fixed correlation with geographical location, serving as a reliable basis for direction determination. It is important to note that this should be consistently described as "rock features" rather than "rock characteristics" to ensure consistency with the technical terminology used earlier. Including rock features as supplementary candidate features allows navigation schemes to still provide usable directional references even in extremely dense forest environments.

[0046] The Texture Enhancement mode is a dedicated acquisition strategy designed specifically for the subtle features of rocks and their susceptibility to background interference. Unlike the spectral focusing mode for plant features and the light and shadow enhancement mode for light and shadow features, it is highly targeted. In Texture Enhancement mode, adjusting the white balance weakens the color interference from soil and decaying leaves because soil and decaying leaves in dense forests are mostly dark brown or yellowish-brown, similar in color to rocks, easily causing texture obscuring. Adjusting the white balance restores the true color tone of the rocks, increasing the color difference between the rocks and surrounding debris, making the boundaries of rock areas clearer. Increasing sharpness and contrast highlights the direction and depth of weathering textures because rock weathering textures are mostly fine grooves or protrusions. Texture features are easily obscured in conventional acquisition modes. Increasing sharpness enhances the details of texture edges, while increasing contrast increases the light and dark differences between texture grooves and the rock surface, making the originally blurry texture direction clearly discernible, thus providing directional reference value. Locking the focus area to the core rock region is crucial to prevent focus drift caused by user-held device movement or surrounding debris. This ensures the collected field of view remains focused on the densest and most stable core area of ​​the rock texture, avoiding the inclusion of irrelevant data from surrounding soil, decaying leaves, and other irrelevant areas, thus guaranteeing sufficient reliability of the collected rock features. These parameter adjustments form a collaboratively optimized acquisition strategy. The ultimate result is clear and minimally disturbed rock feature details in the second set of data, enabling precise extraction of stable directional patterns. This provides high-quality supplementary data for subsequent feature verification and orientation derivation, effectively improving the adaptability and robustness of the entire navigation solution in complex, dense forest environments.

[0047] In some specific embodiments, verifying candidate features based on the second collected data includes: If the candidate feature is a plant feature, the difference in spectral reflectance between the light-facing and shaded sides of the leaves in the second collection data is analyzed. At the same time, the distribution ratio of the leaves in the flat and tilted states is counted. When the reflectance difference has stable distinguishability, or the distribution of leaf orientation shows a consistent trend, and this trend matches the solar orientation pattern, the candidate plant feature is determined to pass the orientation correlation verification. If the candidate feature is a light and shadow feature, then, in conjunction with the solar azimuth interval determined by the local clock, it is verified whether the regional distribution difference of light spot density in the second collection data corresponds to the solar azimuth. At the same time, it is verified whether the extension direction of the shadow of the fixed object is inversely related to the solar azimuth. If at least one of the two verifications satisfies the rule, the candidate feature of light and shadow is determined to have passed the azimuth correlation verification.

[0048] By verifying the orientation correlation, candidate plant features and light and shadow features that truly have geographical orientation are selected, and invalid features that only have morphological attributes but no orientation reference value are eliminated. This provides the core basis for subsequent anti-interference stability verification and final orientation derivation. Its verification logic is completely consistent with the inherent correlation between natural features and the sun's position in the dense forest environment, and has strong pertinence and practicality.

[0049] If the candidate feature is a plant feature, its location correlation verification is carried out from two dimensions: spectrum and morphology. The difference in spectral reflectance between the light-facing and shaded sides of the leaves is analyzed. The core principle is that phototropism in plants leads to higher chlorophyll activity on the light-facing side, resulting in significantly higher light reflectance intensity for specific correlated wavelengths compared to the shaded side. This reflectance difference is not random but exhibits a stable and regular discriminative quality. When this discriminative quality reaches a preset threshold, it means that the plant feature possesses the basic conditions for pointing towards the light source and thus associating with geographical location. Simultaneously, the distribution ratio of leaves in flat and tilted states is statistically analyzed. Flat leaves typically actively face the direction of the light source. The tilted leaves will also show a consistent orientation trend due to their growth habits. For example, most leaves in the same area tilt to the east or south. At this time, it is necessary to match this orientation trend with the solar azimuth pattern calculated based on the local clock. For example, if the sun is in the east during the morning, and the leaves are concentrated in the east, it conforms to the azimuth correlation pattern. This verification step sets two conditions: the stability of reflection difference or the orientation trend matching the solar azimuth. If either condition is met, it can be judged as passing. This flexible judgment logic can adapt to the diversity of plant growth status in dense forest environments and avoid the misjudgment and rejection of effective features due to the strictness of single-dimensional verification.

[0050] If the candidate feature is a light and shadow feature, its directional correlation verification revolves around the spatiotemporal correspondence between light and shadow and the sun's position. First, relying on the local clock of the rugged phone combined with a preset sun position model, the current sun position interval is determined. For example, from 9:00 AM to 11:00 AM, the sun is in the southeast to due south interval, and from 1:00 PM to 3:00 PM, it is in the southwest to due west interval. This is the benchmark for verification. Then, two core correlation points are verified. One is whether the regional distribution difference of light spot density in the second collection data corresponds to the sun's position. The principle is that sunlight will pass through the gaps in the tree canopy to form light spots. Areas with higher light spot density must be consistent with the sun's incident direction. For example, when the sun is in the southeast, the tree canopy faces... More light spots appear on the southeast side, which is a typical correspondence. Secondly, whether the extension direction of the shadow of a fixed object is inversely related to the sun's position is a core principle of light and shadow characteristics pointing to geographical location. The direction of the shadow is always opposite to the direction of the light source; when the sun is to the east, the shadow will inevitably extend westward, and when the sun is to the south, the shadow will inevitably extend northward. This verification step sets two correlation points; satisfying either one is sufficient for a pass. This design can adapt to the complexity of light and shadow distribution in dense forest environments. For example, in some areas, the tree canopy is too dense, resulting in very few light spots but clear shadows, or there are many light spots but the shadows are obscured. Verification can be completed using effective dimensions, avoiding verification failure due to the lack of a single light and shadow pattern. Overall, this directional correlation verification step is not a simple feature screening, but rather a precise identification of candidate features with navigational value based on the scientific correlation between natural features and geographical location. It eliminates non-directional features caused by factors such as messy branches and swaying leaves, and interference from reflective objects, laying a solid foundation for subsequent anti-interference stability verification and final relative location derivation. This ensures that the core basis of the entire navigation scheme is scientifically sound and environmentally adaptable.

[0051] In some specific embodiments, verifying candidate features based on the second acquired data further includes: verifying the stability of candidate features against environmental interference from both temporal and spatial dimensions; extracting continuous multi-frame data from the second acquired data and calculating the fluctuation range of the core indicators of the candidate features; if the fluctuation range is within a preset reasonable range and there are no drastic changes caused by short-term interference, the time dimension verification is passed; dividing the target guidance area into multiple equal sub-regions and extracting the core indicators of candidate features in each sub-region; if the difference between the indicators in each sub-region is within a preset threshold and there is no deviation of the indicators caused by local abnormal interference points, the spatial dimension verification is passed.

[0052] Building upon the orientation correlation verification, a supplementary temporal-spatial dual-dimensional anti-interference stability verification is performed. This further filters out false features caused by short-term dynamic interference and local anomalies in dense forest environments, ensuring that the final determined effective features possess both orientational accuracy and environmental anti-interference capabilities. This provides a stable and reliable core basis for subsequent navigation derivation. This verification logic is precisely designed to address the pain points of complex interference factors and easily distorted features in dense forest environments. Verifying the stability of candidate features against environmental interference from both temporal and spatial dimensions essentially involves a dual check of the persistence and uniformity of candidate features, avoiding misjudgments due to accidental features at a single moment or in a single area, and ensuring that the features possess universality as a navigation basis.

[0053] Extracting the fluctuation amplitude of core indicators of candidate features from multiple consecutive frames of data collected in the second phase is the core operation of time-dimensional verification. These core indicators must correspond to the candidate feature type. For plant features, the core indicators are the chlorophyll-related band reflectance difference value and the leaf orientation distribution ratio; for light and shadow features, the core indicators are light spot density and shadow extension direction angle; for rock features, the core indicator is the texture direction consistency parameter. Dense forest environments contain many short-term interference factors, such as gusts causing leaf swaying, momentary miasma blocking light, and brief branch and leaf swaying blocking shadows. These interferences can cause abnormal fluctuations in the feature indicators of a single frame, but they do not affect the overall pattern of multiple consecutive frames. A preset reasonable range is a feature fluctuation threshold calibrated based on a large amount of measured data from dense forest scenes. When the fluctuation amplitude of the core indicators in multiple consecutive frames is within this range, and there are no drastic jumps caused by short-term interference, the time-dimensional verification is passed. This step aims to eliminate false features caused by instantaneous interference and ensure the stability of candidate features in the time dimension.

[0054] Dividing the target guidance area into multiple equal sub-regions and extracting the core indicators of candidate features from each sub-region is the core operation of spatial dimension verification. The purpose of dividing into equal sub-regions is to achieve full-coverage verification of the target guidance area and avoid the influence of local anomalies on the overall judgment. In dense forest environments, there are local anomalies, such as reflective metallic debris, decaying leaves covering plant leaves, or moss covering the texture of rocks in a certain sub-region. These local anomalies will cause the feature indicators of the corresponding sub-regions to deviate from the normal range, but will not affect the feature regularity of the entire target guidance area. The preset threshold is the upper limit of difference based on measured data of feature spatial distribution uniformity. When the difference of the core indicators of each sub-region is within this threshold and there is no significant deviation of the indicators caused by local anomalies, the spatial dimension verification is deemed successful. The purpose of this step is to eliminate false features caused by local anomalies and ensure that the candidate features are uniform in the spatial dimension.

[0055] The temporal and spatial dimensions must be verified simultaneously to complete the anti-interference stability verification. Combined with the previous orientation correlation verification, a complete verification system of orientation correlation + spatiotemporal stability is formed. Only candidate features that pass the verification at both levels can be finally determined as effective features that are strongly correlated with geographical orientation. This verification logic is progressive and effectively solves the problems of features being easily interfered with and false features being difficult to remove in dense forest environments, which greatly improves the accuracy and reliability of subsequent navigation inference.

[0056] An offline navigation system for intelligent terminals in dense forest environments is shown in the attached diagram. Figure 5 As shown, the navigation system includes: The first acquisition unit 1 is used to perform multimodal scanning of the surrounding dense forest environment through the data acquisition module of the smart terminal to obtain the first acquisition data, and to filter out candidate features related to geographical location from the first acquisition data in combination with the synchronously acquired environmental sensing data. The parsing unit 2 is used to determine the target guidance area that needs to be focused on and the corresponding acquisition mode based on the candidate features, generate a visual acquisition guide based on the target guidance area, and adjust the acquisition parameters of the data acquisition module according to the acquisition mode. The second acquisition unit 3 is used to guide the user to perform multimodal scanning of the target guidance area according to the visual acquisition guide to obtain the second acquisition data, verify candidate features based on the second acquisition data, and take the verified candidate features as effective features that are strongly correlated with geographical location. Fusion unit 4 is used to derive relative orientation based on effective features and output navigation results by combining them with local offline maps.

[0057] Those skilled in the art will understand that the modules described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using computer-executable program code, allowing them to be stored in a storage system for execution by the computing system. Alternatively, they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0058] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

[0059] The above disclosures are only a few specific implementation scenarios of this application. However, this application is not limited to these. Any variations that can be conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. An offline navigation method for intelligent terminals in a dense forest environment, characterized in that, include: The data acquisition module of the smart terminal performs multimodal scanning of the surrounding dense forest environment to obtain the first collection data, and combines the synchronously acquired environmental sensor data to filter out candidate features related to geographical location from the first collection data. Based on the candidate features, the target guidance area that needs to be focused on and the corresponding acquisition mode are determined. A visual acquisition guide is generated based on the target guidance area, and the acquisition parameters of the data acquisition module are adjusted according to the acquisition mode. The user is guided to perform multimodal scanning of the target guidance area according to the visual acquisition guide to obtain the second acquisition data. The candidate features are verified based on the second acquisition data, and the verified candidate features are taken as effective features that are strongly correlated with geographic location. Based on the aforementioned effective features, the relative orientation is derived, and the navigation result is output in conjunction with the local offline map.

2. The offline navigation method for intelligent terminals according to claim 1, characterized in that, The first acquired data includes visual image data and spectral data; the environmental sensing data includes temperature and humidity, air pressure, and ambient light intensity data, and is associated with the first acquired data by timestamp. Based on visual image data and spectral data, potential feature regions are divided, including at least plant areas and light and shadow areas. From each potential feature region, features related to geographical orientation are extracted to obtain candidate features.

3. The offline navigation method for intelligent terminals according to claim 2, characterized in that, For plant areas, spectral data is used to capture the reflectance characteristics of chlorophyll-related bands in leaves, and leaf outline and orientation information are extracted by combining visual images. The plant activity is determined with reference to the temperature and humidity data. When the activity is determined to be high, the priority of this type of plant feature in the candidate feature is increased; when the activity is determined to be low, the priority of this type of plant feature in the candidate feature is decreased. For the light and shadow area, combined with ambient light intensity and air pressure data, regular-shaped light spots and blurry short shadows are excluded, and only irregular sun spots and clear shadows of fixed objects are retained. It is also ensured that the light and shadow distribution initially matches the solar azimuth interval determined based on the local clock to obtain the light and shadow features, and these light and shadow features are included in the candidate features.

4. The offline navigation method for intelligent terminals according to claim 1, characterized in that, Determining the target guidance region based on candidate features specifically includes: Locate the core distribution area of ​​candidate features in the dense forest environment, and delineate the target guidance area based on feature clarity and area ratio to ensure that the target guidance area covers the core part of the candidate features.

5. The offline navigation method for intelligent terminals according to claim 3, characterized in that, If plant features are included in the candidate features, the acquisition mode is switched to spectral focusing mode; In the spectral focusing mode: the hyperspectral acquisition unit is adjusted to focus on the chlorophyll-related band, and interference from other bands is shielded; the camera exposure parameters are optimized, the focus range is fixed to the distance corresponding to the target guidance area, and the white balance and sharpness are adjusted to enhance the leaf outline and texture details, highlighting the difference between the light-facing and shaded sides.

6. The offline navigation method for intelligent terminals according to claim 3, characterized in that, If the candidate features involve light and shadow features, switch the acquisition mode to light and shadow enhancement mode; In the light and shadow enhancement mode: control the camera exposure, switch to spot metering mode to focus on the core area of ​​light and shadow, increase contrast to widen the grayscale difference between the light spot and the background, and between the shadow and the object, and fix the shutter speed to eliminate the light and shadow motion caused by the leaf shaking.

7. The offline navigation method for intelligent terminals according to claim 3, characterized in that, When the light and shadow features and plant features are insufficient, the potential feature region of rock area is divided based on visual image data. Rock features are obtained by extracting the direction information of natural weathering texture on the rock surface and are included as supplementary candidate features. If the candidate features involve rock features, switch the acquisition mode to texture enhancement mode; In the texture enhancement mode: the white balance is adjusted to weaken the color interference of soil and decaying leaves, the sharpness and contrast are improved to highlight the direction and depth of weathering texture, and the focus range is locked to the core area of ​​the rock to ensure clear acquisition of texture details.

8. The offline navigation method for intelligent terminals according to claim 3, characterized in that, Verifying the candidate features based on the second collected data includes: If the candidate feature is a plant feature, the difference in spectral reflectance between the light-facing and shaded sides of the leaves in the second collection data is analyzed. At the same time, the distribution ratio of the leaves in the flat and tilted states is counted. When the reflectance difference has stable distinguishability, or the distribution of leaf orientation shows a consistent trend, and this trend matches the solar orientation pattern, the candidate plant feature is determined to pass the orientation correlation verification. If the candidate feature is a light and shadow feature, then, in conjunction with the solar azimuth interval determined by the local clock, it is verified whether the regional distribution difference of light spot density in the second collection data corresponds to the solar azimuth. At the same time, it is verified whether the extension direction of the shadow of the fixed object is inversely related to the solar azimuth. If at least one of the two verifications satisfies the rule, the candidate feature of light and shadow is determined to have passed the azimuth correlation verification.

9. The offline navigation method for intelligent terminals according to claim 8, characterized in that, Verifying the candidate features based on the second collected data further includes: The stability of candidate features against environmental interference was verified from both temporal and spatial dimensions. Extract multiple consecutive frames of data from the second collection data, calculate the fluctuation range of the core indicators of the candidate features, and if the fluctuation range is within a preset reasonable range and there is no drastic change caused by short-term interference, it is determined to pass the time dimension verification. The target guidance area is divided into multiple equal sub-regions, and the core indicators of candidate features are extracted from each sub-region. If the difference between the indicators of each sub-region is within the preset threshold and there is no deviation of the indicators caused by local abnormal interference points, it is determined to pass the spatial dimension verification.

10. An offline navigation system for intelligent terminals in a dense forest environment, characterized in that, include: The first acquisition unit is used to perform multimodal scanning of the surrounding dense forest environment through the data acquisition module of the smart terminal to obtain the first acquisition data, and to filter out candidate features related to geographical location from the first acquisition data in combination with the synchronously acquired environmental sensing data. The parsing unit is used to determine the target guidance area that needs to be focused on and the corresponding acquisition mode based on the candidate features, generate a visual acquisition guide based on the target guidance area, and adjust the acquisition parameters of the data acquisition module according to the acquisition mode. The second acquisition unit is used to guide the user to perform multimodal scanning of the target guidance area according to the visual acquisition guide to obtain the second acquisition data, verify the candidate features based on the second acquisition data, and take the verified candidate features as effective features that are strongly correlated with geographical location. The fusion unit is used to deduce the relative orientation based on the effective features and output navigation results by combining them with the local offline map.

Images