Picture viewing method and device based on intelligent pod and electronic equipment

Abstract

The invention provides a picture viewing method and device based on an intelligent pod and electronic equipment, and relates to the field of image processing, and the method comprises the steps: receiving a first viewing instruction, the first viewing instruction is determined after a user selects a local picture viewing angle in a panoramic picture, and the panoramic picture is an original picture obtained through a multi-path camera, and is determined according to the local picture viewing angle; splicing the original pictures to determine a first resolution picture; displaying a current display picture according to the local picture view angle, determining a preloaded picture according to the current display picture and a preset magnification factor, and replacing the target recognition object corresponding to the first resolution with a target recognition object corresponding to the second resolution in the preloaded picture; and displaying a target picture with a target recognition object corresponding to a second resolution, determining the target picture in the preloaded pictures, and displaying the target picture in a preset display interface. According to the invention, local high-definition display of the returned data of the intelligent pod is realized through a real-time interactive preloading strategy.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a method, device and electronic equipment for viewing pictures based on a smart pod. Background Art

[0002] The smart pod seamlessly stitches images collected by multiple cameras into a complete panoramic image, and transmits the panoramic image to the ground control and display equipment in real time. Due to the limited size of the ground display equipment, only one angle of the panoramic image can be seen at the same time. Therefore, the smart pod system reads the panoramic image viewing angle required by the ground control and display equipment in real time, and crops the stitched panoramic image in real time. It can transmit back part or all of the panoramic images within the viewing angle of the ground control and display equipment. The ground control and display equipment can use the mouse to select the area of ​​interest on the displayed panoramic image, and achieve local high definition by clicking the area of ​​interest.

[0003] Since all interactive images currently need to be transmitted back from the smart pod to the ground control and display devices, this processing method will undoubtedly lead to an increase in the amount of real-time transmission data when the user switches the perspective to interact. In addition, the frequent changes and adjustments required to display the local high-definition will cause the local high-definition loading to be stuck, delayed or fail, which will not only increase the unnecessary waste of communication resources, but also reduce the real-time and stability of the entire smart pod system.

[0004] However, there is currently no technical solution that can solve the above technical problems, and there is no image viewing method, device and electronic equipment based on a smart pod. Summary of the invention

[0005] The present invention provides a method, device and electronic equipment for viewing pictures based on a smart pod, which realize local high-definition display of data returned by the smart pod through a real-time interactive preloading strategy.

[0006] In a first aspect, the present invention provides a method for viewing pictures based on a smart pod, comprising:

[0007] Receive a first viewing instruction from a user, where the first viewing instruction is determined based on a partial image perspective selected by the user in the panoramic image returned by the smart pod, where the panoramic image is a first resolution image determined by splicing all the original images after obtaining the original images corresponding to each channel using multiple cameras of the smart pod;

[0008] In response to the first viewing instruction, a currently displayed image is displayed in a preset display interface according to the local image viewing angle, a preloaded image is determined according to the currently displayed image and a preset magnification, and when it is determined that a target recognition object exists in the preloaded image, a target recognition object corresponding to a first resolution is replaced in the preloaded image with a target recognition object corresponding to a second resolution, the target recognition object corresponding to the second resolution is obtained after the intelligent pod performs resolution enhancement on the panoramic image, and is determined after transmitting back the position of the target recognition object in the enhanced image, the target recognition object is determined after the intelligent pod performs target recognition on the enhanced image using a preset target recognition model, and the first resolution is smaller than the second resolution;

[0009] Receive and respond to a second viewing instruction from the user, where the second viewing instruction is used to instruct display of a target image with a target recognition object corresponding to the second resolution, determine the target image from the preloaded images, and display the target image in a preset display interface.

[0010] According to the picture viewing method based on the smart pod provided by the present invention, before receiving the first viewing instruction from the user, the method further includes:

[0011] The intelligent pod uses multiple cameras to collect images from all angles within the hemisphere to obtain the original images corresponding to each channel. The multiple cameras include a camera at the top of the hemisphere, four cameras corresponding to the middle of the hemisphere, and four cameras corresponding to the horizontal position of the hemisphere.

[0012] According to the plane coordinates corresponding to each camera, each corresponding original image is mapped to the longitude and latitude coordinates corresponding to the hemisphere, and the hemispherical panoramic image of the first resolution is output.

[0013] According to the picture viewing method based on the smart pod provided by the present invention, before responding to the viewing instruction, the method further includes:

[0014] For each original picture in the panoramic picture, when the brightness of any original picture is lower than a preset brightness, input the original picture into a low illumination enhancement network model to obtain an illumination enhanced picture output by the low illumination enhancement network model;

[0015] Traversing all original pictures to determine all pictures to be processed, where the pictures to be processed include original pictures and / or illumination enhanced pictures;

[0016] The low illumination enhancement network model is determined after training based on all low illumination sample pictures and illumination enhancement sample pictures corresponding to each low illumination sample picture.

[0017] According to the picture viewing method based on the smart pod provided by the present invention, after all pictures to be processed are determined, the method further includes:

[0018] Input each image to be processed into a super-resolution enhancement network model, and obtain each resolution-enhanced image output by the super-resolution enhancement network model;

[0019] splicing all the resolution-enhanced pictures to obtain the enhanced picture of the second resolution;

[0020] The super-resolution enhancement network model is determined after training based on all sample processed images and the sample enhanced images corresponding to each sample processed image.

[0021] According to the picture viewing method based on the smart pod provided by the present invention, after obtaining the enhanced picture, the method further includes:

[0022] Using a preset step length as a sliding speed, cropping the enhanced image using a preset slider window to obtain all cropped images;

[0023] For each cropped image, input the cropped image to a preset target recognition model, and obtain a target recognition result output by the preset target recognition model, wherein the target recognition result includes the presence of the target recognition object and the absence of the target recognition object;

[0024] The preset target recognition model is determined after training based on all sample recognition images containing target recognition objects and sample recognition images not containing target recognition objects.

[0025] According to the picture viewing method based on the smart pod provided by the present invention, before obtaining all the cropped pictures, the method further includes:

[0026] According to the recognition type of the target recognition object, determining the window size value of the preset slider window from the corresponding relationship between the preset window size and the recognition type;

[0027] According to the recognition type of the target recognition object, the step value of the preset step size is determined from the corresponding relationship between the preset step size and the recognition type.

[0028] According to the picture viewing method based on the smart pod provided by the present invention, after determining the preloaded picture according to the currently displayed picture and the preset magnification, the method further includes:

[0029] When it is determined that all target recognition objects exist in the preloaded image, replacing the target recognition objects corresponding to the first resolution with the target recognition objects corresponding to the second resolution in the preloaded image;

[0030] When it is determined that some target recognition objects exist in the preloaded image, the preset magnification is increased, a preloaded image corresponding to all target recognition objects is determined, and the target recognition objects corresponding to the first resolution are replaced with the target recognition objects corresponding to the second resolution in the preloaded image.

[0031] According to the picture viewing method based on the smart pod provided by the present invention, after receiving and responding to the second viewing instruction of the user, the method further includes:

[0032] At every preset time interval, the updated panoramic picture returned by the smart pod is updated in the preset display interface, and again in response to the first viewing instruction, the updated display picture is displayed in the preset display interface according to the local picture perspective, and again the second viewing instruction of the user is received and responded to, and the updated picture with the target recognition object corresponding to the second resolution is displayed.

[0033] In a second aspect, the present invention provides a picture viewing device based on a smart pod, comprising:

[0034] A first receiving unit, the first receiving unit is used to receive a first viewing instruction of a user, the first viewing instruction is determined after the user selects a local image perspective in the panoramic image returned by the smart pod, the panoramic image is a first resolution image determined after the original image corresponding to each channel is obtained by using multiple cameras of the smart pod and all the original images are stitched together;

[0035] a response unit, the response unit being used to respond to the first viewing instruction, display the current display image in a preset display interface according to the local image viewing angle, determine the preloaded image according to the current display image and the preset magnification, and replace the first resolution corresponding target recognition object with the second resolution corresponding target recognition object in the preloaded image when it is determined that there is a target recognition object in the preloaded image, the second resolution corresponding target recognition object is obtained after the intelligent pod performs resolution enhancement on the panoramic image, and is determined after transmitting back according to the position of the target recognition object in the enhanced image, the target recognition object is determined after the intelligent pod performs target recognition on the enhanced image using a preset target recognition model, and the first resolution is smaller than the second resolution;

[0036] A second receiving unit, the second receiving unit is used to receive and respond to a second viewing instruction of the user, the second viewing instruction is used to instruct the display of a target image with a target recognition object corresponding to the second resolution, determine the target image in the preloaded images, and display the target image in a preset display interface.

[0037] In a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the smart pod-based image viewing method when executing the program.

[0038] When performing panoramic picture preview, the present invention uses a lower-resolution picture to search for layout target recognition objects. When it is determined that the target recognition object exists, the high-resolution layout picture is transmitted back to cover the previous lower-resolution picture. By presetting the local high-definition function, the user is allowed to perform high-definition processing on specific areas according to actual needs, further improving the monitoring quality of key areas, reducing the requirements for network bandwidth, and meeting diversified and personalized needs. The partial panoramic picture transmission technology not only ensures the real-time performance of the data, but also effectively reduces the dependence on network bandwidth, so that the system can work stably in an environment with poor network conditions, meeting the application scenarios with high real-time requirements;

[0039] The low-light enhanced deep network carried by the present invention enables the system to clearly capture and identify targets at night or in dim light conditions, ensuring efficient operation around the clock and at all times. Combined with super-resolution deep network technology, the system can accurately identify targets at long distances and retain the fine details of the targets. Even if the targets are at the edge of the field of vision or at a farther position, high-quality recognition can be achieved. It shows significant advantages in functional diversity, all-weather adaptability, long-distance recognition capability, local high-definition function, recognition accuracy, panoramic field of view coverage, and real-time data transmission. It can better adapt to complex and changeable field conditions and complex task requirements, and provide users with more comprehensive, efficient, and personalized intelligent monitoring solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 It is a flow chart of a method for viewing pictures based on a smart pod provided by the present invention;

[0042] Figure 2 It is a structural schematic diagram of a picture viewing device based on a smart pod provided by the present invention;

[0043] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Figure 1 : is a flow chart of a method for viewing pictures based on a smart pod provided by the present invention, wherein the method for viewing pictures based on a smart pod comprises:

[0046] Step 101: receiving a first viewing instruction from a user, wherein the first viewing instruction is determined by selecting a local image perspective from a panoramic image transmitted back by the smart pod according to the user, wherein the panoramic image is a first resolution image determined by splicing all the original images after obtaining the original images corresponding to each channel using multiple cameras of the smart pod;

[0047] Step 102: In response to the first viewing instruction, the currently displayed image is displayed in a preset display interface according to the local image viewing angle, a preloaded image is determined according to the currently displayed image and a preset magnification, and when it is determined that a target recognition object exists in the preloaded image, the target recognition object corresponding to the first resolution is replaced in the preloaded image with a target recognition object corresponding to the second resolution, the target recognition object corresponding to the second resolution is obtained after the intelligent pod performs resolution enhancement on the panoramic image, and is determined after transmitting back the position of the target recognition object in the enhanced image, the target recognition object is determined after the intelligent pod performs target recognition on the enhanced image using a preset target recognition model, and the first resolution is smaller than the second resolution;

[0048] Step 103, receiving and responding to a second viewing instruction from the user, wherein the second viewing instruction is used to instruct displaying a target image with a target recognition object corresponding to the second resolution, determining the target image in the preloaded images, and displaying the target image in a preset display interface.

[0049] In step 101, the executor of the present invention is a ground control and display device, which mainly utilizes the multi-channel cameras of the smart pod to obtain panoramic images, and displays local high-definition images according to the user's viewing instructions. The system includes a smart pod, a ground control and display device, and an autonomous network connecting the two. The user selects a local image perspective from the panoramic image sent back by the smart pod through the ground control and display device, and determines a first viewing instruction. The panoramic image is a first resolution image determined by obtaining the original image corresponding to each channel using the multi-channel cameras of the smart pod and stitching them together using a preset stitching algorithm. The first resolution image is set to a relatively low resolution and is only provided to the user for preview viewing.

[0050] Optionally, before receiving the first viewing instruction from the user, the method further includes:

[0051] The intelligent pod uses multiple cameras to collect images from all angles within the hemisphere to obtain the original images corresponding to each channel. The multiple cameras include a camera at the top of the hemisphere, four cameras corresponding to the middle of the hemisphere, and four cameras corresponding to the horizontal position of the hemisphere.

[0052] According to the plane coordinates corresponding to each camera, each corresponding original image is mapped to the longitude and latitude coordinates corresponding to the hemisphere, and the hemispherical panoramic image of the first resolution is output.

[0053] Optionally, the smart pod is equipped with multiple cameras, including a camera on the top of the hemisphere to capture the view above; four cameras corresponding to the middle position of the hemisphere to capture the view around the middle section; four cameras corresponding to the horizontal position of the hemisphere, the goal is to collect image data from all angles within the hemisphere, and then use the USB interface to connect the nine cameras to the edge computing module, transfer the collected image data and wait for the next step of processing, which is used to capture the ground and horizontal field of view, ensuring that images of all angles within the hemisphere can be collected, each camera is responsible for collecting images within its corresponding field of view, generating original pictures, and setting corresponding Plane coordinates, and pre-set the latitude and longitude coordinate system corresponding to the hemisphere. According to the plane coordinates corresponding to each camera, the preset image stitching algorithm and mapping rules are used to map each original image to the latitude and longitude coordinates corresponding to the hemisphere, ensuring seamless stitching between images, forming a complete and continuous hemispherical panoramic image. According to actual application requirements and transmission bandwidth limitations, the first resolution of the output panoramic image is set. The selection of the first resolution should minimize the size of the image file while ensuring image clarity to improve transmission efficiency. After mapping processing, a hemispherical panoramic image is obtained, and it is output with the first resolution set for subsequent viewing and processing. The present invention realizes image acquisition and stitching of all angles within the hemisphere through the layout of multiple cameras and image mapping technology, forming a complete hemispherical panoramic image, greatly expanding the monitoring field of view, enabling the system to adapt to the needs of various complex scenes, and realizing all-round and blind-angle-free monitoring. In the image mapping process, the transmission efficiency is also improved through a smaller image file size, reducing the dependence on network bandwidth.

[0054] Optionally, the present invention can also integrate multi-channel visible light detectors, support frame synchronization and dynamic white balance functions among multiple cameras, and reduce the acquisition width of a single-channel sensor lens by utilizing the quantity advantage of multi-channel sensors, thereby improving the ground observation resolution of a single-channel sensor and implementing a distributed structure layout to achieve the real-time acquisition capability of panoramic wide-area motion images greater than 360°. Two infrared cameras are provided to simultaneously collect infrared data for the area of ​​interest, which is subsequently compared and matched with the panoramic data and the data is fused.

[0055] Optionally, the plane image data corresponding to each camera is mapped to the corresponding hemispherical fixed points according to the set fixed key points. After the key points are fixed, the coordinates of other points on the image can be mapped to the hemisphere one by one according to similar rules, which is equivalent to making a mapping from plane coordinates to hemispherical longitude and latitude coordinates for each point on the plane. Finally, a hemispherical panoramic image is output for ground control and display equipment to call and view. The ground control / display equipment transmits the panoramic picture perspective to be called and viewed to the above-mentioned smart pod system (multi-channel camera + edge computing module). After receiving the signal, the smart pod system returns the partial panoramic picture required by the ground. In the present invention, the intelligent pod system and the ground control / display device are connected and communicated through the autonomous network hardware device. Since the size of the ground display device is limited, only one angle of the panoramic picture can be seen at the same time. Therefore, the intelligent pod system reads the panoramic picture viewing angle required by the ground control / display device in real time, and crops and splices the panoramic picture in real time, and only transmits back part of the panoramic picture within the viewing angle of the ground control / display device, avoiding the burden on communication transmission caused by the excessive size of the entire panoramic picture, avoiding unnecessary waste of communication resources, and improving the real-time and stability of the entire intelligent pod system.

[0056] In step 102, in the preset display interface, the currently displayed image is displayed according to the local image perspective, and the preloaded image range is determined according to the currently displayed image and the preset magnification. The preloaded image range can be the surrounding area of ​​the current image perspective to ensure that the user can quickly load the high-definition image when switching the perspective. When it is determined that there is a target recognition object in the preloaded image, the target recognition object in the preloaded image is locally processed for high definition. The local high-definition processing includes: the smart pod performs resolution enhancement on the panoramic image to obtain an enhanced image; performs target recognition on the enhanced image using a preset target recognition model to determine the target recognition object; and transmits back the position of the target recognition object in the enhanced image to obtain a target recognition object corresponding to a second resolution, wherein the second resolution is higher than the first resolution, and the target recognition object corresponding to the second resolution is replaced with the corresponding position in the preloaded image, thereby realizing the data transmission of the target recognition object corresponding to the second resolution.

[0057] Optionally, after receiving the command, the smart pod system processes the high-definition original image of the specified local area into a map and transmits it back to the ground. For example, a lower downsampling level is selected or the original image is directly transmitted back to cover the corresponding position of the low-sampled base map. After obtaining the target image perspective on the ground, the surrounding field of view 1.5 times around it is preloaded without affecting the transmission rate of the target perspective. For example, an image with the same clarity as the target perspective or one level of downsampling is loaded in advance as a preloaded image. The preloading function ensures that when the user needs to view other perspectives after completing the current call to view the command, the system can respond quickly based on preloading.

[0058] Optionally, before responding to the viewing instruction, the method further includes:

[0059] For each original picture in the panoramic picture, when the brightness of any original picture is lower than a preset brightness, input the original picture into a low illumination enhancement network model to obtain an illumination enhanced picture output by the low illumination enhancement network model;

[0060] Traversing all original pictures to determine all pictures to be processed, where the pictures to be processed include original pictures and / or illumination enhanced pictures;

[0061] The low illumination enhancement network model is determined after training based on all low illumination sample pictures and illumination enhancement sample pictures corresponding to each low illumination sample picture.

[0062] Optionally, in the panoramic smart pod system, for each original image in the panoramic image, the system will first perform a brightness judgment. If the brightness of an original image is lower than a preset brightness threshold, the image is regarded as a low-light image and needs to be enhanced in illumination. Specifically, the original image with a brightness lower than the preset threshold is input into a low-light enhancement network model. The low-light enhancement network model is an algorithm based on deep learning. The model has learned how to restore high-quality, illumination-enhanced images from low-light images during the training phase. The training data includes a large number of low-light sample images and their corresponding illumination-enhanced sample images. The low-light enhancement network model is used to process the image and output the image with enhanced illumination, which significantly improves the brightness while maintaining the naturalness and clarity of the image. After completing the enhancement processing of the low-light image, the system will traverse all the original images and determine all the images that need subsequent processing. These images are called images to be processed. The images to be processed may include original images with sufficient brightness, images that have been processed with illumination enhancement, or both. For the images to be processed, the system will continue to perform subsequent steps such as super-resolution enhancement, target recognition and positioning, panoramic image stitching, and local high-definition processing.

[0063] Optionally, the preset logic carried in the edge computing module determines whether the brightness of the currently incoming image data is lower than a preset threshold value. If the image data incoming from a certain camera is lower than the preset threshold value, it is input into the low-light enhancement deep network for enhancement processing. The image data transmitted to the edge computing module by the nine cameras in the previous step is first subjected to logical judgment. If the brightness of the image data incoming from a certain camera is lower than the preset threshold value, the image data is input into the low-light enhancement deep network for enhancement processing. The low-light enhancement network used here is the supervised learning low-light enhancement algorithm SMG-LLIE. Through the processing of the low-light enhancement network model, the present invention can significantly enhance the brightness of the original picture taken under low-light conditions, thereby improving the overall quality of the image. This enables the panoramic intelligent pod system to capture clear and high-quality images at night or in low-light conditions, enhancing the system's all-weather adaptability. The brightness-enhanced picture can provide clearer and more accurate feature information in the subsequent target recognition and positioning steps, which helps to improve the accuracy and reliability of target recognition, especially in complex and changing environments.

[0064] Optionally, after all the pictures to be processed are determined, the method further includes:

[0065] Input each image to be processed into a super-resolution enhancement network model, and obtain each resolution-enhanced image output by the super-resolution enhancement network model;

[0066] splicing all the resolution-enhanced pictures to obtain the enhanced picture of the second resolution;

[0067] The super-resolution enhancement network model is determined after training based on all sample processed images and the sample enhanced images corresponding to each sample processed image.

[0068] Optionally, after determining all the images to be processed, these images are input into the super-resolution enhancement network model one by one. The super-resolution enhancement network model is a model based on deep learning. It has been trained on a large number of sample processing images and their corresponding sample enhancement images. The model can learn and extract detailed information in the image, and perform efficient super-resolution enhancement processing on the image by combining intelligent algorithms and graphics processing accelerators. Specifically, first establish a data set, use a public low-light data set (such as LOL, SID, etc.) or a collected custom low-light data set. Take the collected custom low-light data set as an example, and pre-process it for training and verifying the SMG-LLIE model; in model training, use the pre-processed custom low-light data set to train the SMG-LLIE model. During the training process, the pre-trained model can be used for transfer learning to accelerate the training process and improve the model performance; in model verification and optimization, evaluate the model performance on the verification data set, calculate indicators such as ranging error, and if necessary, optimize the model, such as adjusting the network structure, loss function or optimizer, etc.; in deployment, deploy the trained SMG-LLIE model to the edge computing module. During the deployment process, it is necessary to read the image data of nine cameras in real time, and then use the SMG-LLIE model to perform low-light enhancement processing on the image data that does not reach the brightness threshold. The processed image data corresponding to the nine cameras are respectively input into the super-resolution enhancement deep network to enhance the edge of the target in the image. After the above processing, the image data corresponding to all nine cameras has a brightness higher than the preset brightness, and then it is input into the super-resolution enhancement deep network for enhancement processing to enhance the edge information of the target in the image, so as to facilitate the target recognition operation of the next target recognition network. The super-resolution enhancement algorithm used is EDSR, and the training data set is DIV2K.

[0069] Optionally, after being processed by the super-resolution enhancement network model, each image to be processed will output a picture with significantly enhanced resolution, i.e., a resolution-enhanced picture. All resolution-enhanced pictures are spliced ​​to form a complete and resolution-enhanced panoramic picture, i.e., the enhanced picture corresponding to the second resolution. Through the processing of the super-resolution enhancement network model, the resolution of each part of the panoramic picture is significantly improved, which enables the system to capture high-quality panoramic pictures in various complex environments, and provides strong support for the subsequent target recognition of target objects.

[0070] As another optional embodiment, the present invention can also first determine whether the brightness is lower than a certain threshold, and use a low-light enhancement deep network to enhance the image with lower brightness; input the image with enhanced brightness into a super-resolution enhancement deep network to enhance the contour of the target in the image; input the image with enhanced contour and brightness into a recognition deep network to obtain the position coordinates of the target in the image and an identification frame; map the processed plane image to hemispherical longitude and latitude coordinates according to pre-set key points to obtain a spliced ​​panoramic image; finally, the intelligent pod obtains the panoramic picture range that the ground control and display equipment needs to view through the autonomous network, calculates the viewing angle, crops the panoramic picture, and then transmits it back to the ground control and display equipment through the autonomous network.

[0071] Optionally, after obtaining the enhanced image, the method further includes:

[0072] Using a preset step length as a sliding speed, cropping the enhanced image using a preset slider window to obtain all cropped images;

[0073] For each cropped image, input the cropped image to a preset target recognition model, and obtain a target recognition result output by the preset target recognition model, wherein the target recognition result includes the presence of the target recognition object and the absence of the target recognition object;

[0074] The preset target recognition model is determined after training based on all sample recognition images containing target recognition objects and sample recognition images not containing target recognition objects.

[0075] Optionally, based on the preliminary detection, a slider window recognition method is used to perform a more detailed local scan and analysis of the image. A slider window of a fixed size is set, and the window slides on the image according to a preset step size. Each time the window slides to a new position, the image area in the window is input into the target recognition network for local target detection. By comparing the detection results in different windows, more smaller or occluded targets that may be missed by global detection can be identified. Combining the global and local detection results, the identified targets are fused and deduplicated to ensure that each target is identified only once, and the precise coordinate position and identification frame of the target are output. The introduction of the slider window recognition method not only improves the recognition rate of long-distance targets, but also enhances the system's ability to handle complex backgrounds, occluded or overlapping targets, making the results of target recognition and positioning more accurate and reliable.

[0076] Through the above steps, the smart pod system can perform in-depth local analysis of every detail in the image while maintaining global monitoring, ensuring accurate identification and positioning of targets even at long distances or in complex scenarios.

[0077] Optionally, after obtaining the enhanced image, first preset a step size as the sliding speed, this step size determines the speed at which the slider window slides on the image, and uses the preset slider window to crop the enhanced image. The slider window will slide on the image according to the preset step size, and each slide will crop a new image area. In this way, all cropped images can be obtained, which contain different areas in the enhanced image. For each cropped image, it is input into a preset target recognition model. The preset target recognition model is a model based on deep learning, which has been used in a large number of sample recognition images containing target recognition objects and samples without target recognition objects. The model is trained on sample recognition images of objects. It can automatically extract features from images and determine whether there is a target recognition object in the image based on these features. The target recognition model will output a target recognition result, which includes two cases: the presence of a target recognition object and the absence of a target recognition object. The cropped image is more focused on the local area, which helps the target recognition model to more accurately extract and recognize the features of the target object. By cropping the enhanced image and using the preset target recognition model for target recognition, the efficiency and accuracy of target recognition can be significantly improved, the consumption of computing resources can be reduced, the application requirements of different scenarios can be adapted, and the overall performance of the system can be improved.

[0078] Optionally, before obtaining all cropped images, the method further includes:

[0079] According to the recognition type of the target recognition object, determining the window size value of the preset slider window from the corresponding relationship between the preset window size and the recognition type;

[0080] According to the recognition type of the target recognition object, the step value of the preset step size is determined from the corresponding relationship between the preset step size and the recognition type.

[0081] Optionally, according to different target categories, the present invention trains a set of adaptive slider window clipping methods, and can determine under which slider window parameters (size and overlap ratio) the recognition rate of the target category is the highest according to the average size of the recognition frame of the target class mark (such as calculated by the number of pixels) according to network training. In another embodiment, according to the recognition type of the target recognition object, the system will first search for a correspondence table between a preset window size and a recognition type. In this correspondence, different recognition types will correspond to different window size values. These window size values ​​are determined comprehensively based on factors such as the characteristics, size, and image resolution of the target recognition object. The system will determine a suitable window size value from the correspondence according to the current target recognition object type as the window size of the preset slider window; similar to determining the window size value, the system will also search for a preset step size and a correspondence table between the recognition type according to the recognition type of the target recognition object. In this correspondence, different recognition types will correspond to different step values. These step values ​​are determined comprehensively based on factors such as the distribution density, image complexity, and computing resources of the target recognition object. According to the current target recognition object type, a suitable step value is determined from the correspondence as the step value of the preset step. After determining the window size and step value, the system can use the preset slider window to crop the enhanced image. In this way, all cropped images can be obtained, which contain different areas in the enhanced image, and the size and cropping interval of each area are optimized according to the characteristics and requirements of the target recognition object. The present invention can dynamically adjust the window size and step value according to different target recognition object types, and can be flexibly applied to different scenarios and fields to adapt to different target recognition needs.

[0082] Optionally, after determining the preloaded image according to the currently displayed image and the preset magnification, the method further includes:

[0083] When it is determined that all target recognition objects exist in the preloaded image, replacing the target recognition objects corresponding to the first resolution with the target recognition objects corresponding to the second resolution in the preloaded image;

[0084] When it is determined that some target recognition objects exist in the preloaded image, the preset magnification is increased, a preloaded image corresponding to all target recognition objects is determined, and the target recognition objects corresponding to the first resolution are replaced with the target recognition objects corresponding to the second resolution in the preloaded image.

[0085] Optionally, after determining the preloaded image, the system will detect the target recognition objects thereon. If all the target recognition objects exist in the preloaded image, the system will replace these target objects from the target recognition objects corresponding to the first resolution with the target recognition objects corresponding to the second resolution. If only some of the target recognition objects exist in the preloaded image, the system will increase the preset magnification and re-determine the preloaded image corresponding to all the target recognition objects. This process may be repeated until a preloaded image containing all the target recognition objects and having a resolution that meets the requirements is found, and then the target recognition objects are replaced from the first resolution with the target recognition objects corresponding to the second resolution. The present invention can dynamically adjust the preset magnification and re-determine the preloaded image according to actual conditions to adapt to different complete display requirements, which enables the system to flexibly respond to various complex scenes and changes.

[0086] Optionally, after receiving and responding to the second viewing instruction from the user, the method further includes:

[0087] At every preset time interval, the updated panoramic picture returned by the smart pod is updated in the preset display interface, and again in response to the first viewing instruction, the updated display picture is displayed in the preset display interface according to the local picture perspective, and again the second viewing instruction of the user is received and responded to, and the updated picture with the target recognition object corresponding to the second resolution is displayed.

[0088] Optionally, when the user issues a second viewing instruction through a certain interactive method, which may be clicking a button or a voice command, the system will receive and respond to the instruction. The second viewing instruction generally indicates that the user wishes to view the target recognition object again with higher resolution or more detailed information. After receiving the second viewing instruction, the system will start a timed task to obtain the latest updated panoramic picture from the smart pod at preset intervals (such as 5 seconds, 10 seconds, etc., specifically set according to system performance and user needs). The obtained updated panoramic picture will be immediately updated in the preset display interface so that the user can see the latest panoramic information in real time. While updating the panoramic picture, if the user issues the first viewing instruction again, the system will respond to the first viewing instruction and display the corresponding partial picture in the preset display interface according to the partial picture perspective specified by the user or the user's previous selection. While displaying the partial picture, if the user issues the second viewing instruction again, the system will search for and display the updated picture with the corresponding target recognition object at the second resolution.

[0089] In the present invention, if the preset time interval is set to be short, it is actually possible to view the relevant pictures of the smart pod based on the video stream, thereby improving the user experience; by continuously adjusting and updating the panoramic pictures acquired in real time, the user is provided with clearer and more accurate target information. In such an embodiment, the data generated by the data acquisition sensor matrix is ​​panoramically stitched in real time to form a panoramic wide area motion image (WAMI) data stream. Based on the high-precision data obtained by the positioning sensor and the attitude sensor, the WAMI data stream is spatially corrected frame by frame to ensure that the image keeps the same coordinates of the same point during the movement of the mobile platform. After stitching and spatial correction, the image is modeled in real time by a visual neural network (VNN) to form a region of interest (ROI), which is accessed in real time according to the resolution and range type, providing local panoramic data storage function, remote transmission function in a broadband network, and medium-bandwidth ROI real-time data stream push function.

[0090] Optionally, the present invention provides a multi-spectral fusion function for different data types generated by multi-source heterogeneous data sources (RGB-T), performs real-time precise coordinate matching and fusion, overlays other non-rectangular area data in real time on the basis of WAMI data, and provides a subsequent ROI unified access interface to enhance the information volume of the same-name points of the WAMI data stream for irregular distortion data generated across time and across means and devices, provides RGB-T data fusion for dim and weak scenes of interest, realizes automatic stretching of the color space of images under saturated and unsaturated lighting conditions, amplifies subtle photon differences in multi-bit width raw data, and supports real-time automatic enhancement of terrain and camouflaged targets.

[0091] Furthermore, based on the data stream after image fusion processing, the interpretable data of different geo-elements and pixel resolutions are read through real-time ROI, and AI / ML technology is used in combination with pre-training and preset threat model data for real-time reasoning to achieve the discovery of potential threats and target identification, generate structured information, and comprehensively determine the coordinate information of the identified target, ultimately forming structured intelligence information and providing intelligence tailoring-aided decision-making information based on target identity.

[0092] The present invention also provides a design and implementation method of a panoramic intelligent pod system that meets local high-definition and real-time transmission, including image acquisition and recognition functions of multiple cameras, and the specific implementation steps are as follows:

[0093] First, multiple cameras are used to collect images. The system first determines whether the brightness of each frame of the image is lower than the preset threshold. For images with lower brightness, a low-light enhancement deep network is used for enhancement processing to improve the brightness performance of the image; the image after brightness enhancement processing is then input into the super-resolution enhancement deep network to further sharpen the contours of the target in the image and improve the detail clarity of the image; the image with enhanced contours and brightness is input into the recognition deep network to identify the target in the image, and obtain the position coordinates of the target and the corresponding recognition frame; the processed plane image is mapped to the hemispherical longitude and latitude coordinates according to the key points set in advance, and a complete panoramic image is obtained through the preset stitching algorithm; the ground control and display equipment can use the mouse to display the panoramic image The area of ​​interest is selected as shown above. After receiving this instruction, the system will send back a local high-definition picture of the selected area to ensure that the image details of the area are clearer; after identifying the target, the smart pod automatically sends back a local high-definition picture of the target and its surrounding area. This function is based on the results of the recognition deep network and can respond in real time and highlight key targets and their surroundings; the smart pod receives the panoramic image range that the ground control and display equipment needs to view through the autonomous network, calculates the corresponding viewing angle, and crops the panoramic image according to this viewing angle; the cropped and partially high-definition processed images are transmitted back to the ground control and display equipment in real time through the autonomous network, ensuring that ground operators can obtain clear and accurate panoramic and local high-definition image information in real time.

[0094] Those skilled in the art understand that panoramic stitching technology is to seamlessly stitch images collected by multiple cameras into a complete panoramic image. This function is crucial for comprehensive monitoring and situational awareness; super-resolution enhancement is to improve image resolution through a preset algorithm, especially to enhance the contour details of the target in the image, which is particularly critical for long-distance target recognition; image recognition and target positioning is to intelligently identify the processed image, accurately output the position coordinates of the target in the image and its identification frame, and provide strong support for subsequent decision-making and actions. However, most of the existing pod technologies only cover one to three of the above functions, resulting in their versatility and adaptability when facing complex and changeable field conditions or diversified needs. The present invention is seriously limited in its real-time performance. According to the specific needs of the ground control and display equipment, the present invention intelligently calculates the range of the panoramic picture to be viewed, and accurately crops the panoramic picture accordingly. The cropped image is efficiently transmitted through the autonomous network, which effectively solves the transmission delay and lack of real-time performance of the panoramic picture caused by its large size. According to the specific needs of the ground control and display equipment, the present invention uses a preloading method to transmit high-definition maps of the vicinity of intelligently identified targets and customized local areas to the ground, while ensuring the information integrity and accuracy of the returned pictures in a low-speed network environment. This innovation not only improves the comprehensive performance of the pod system, but also significantly enhances its applicability and practicality in complex environments and diversified tasks.

[0095] When performing panoramic picture preview, the present invention uses a lower-resolution picture to search for layout target recognition objects. When it is determined that the target recognition object exists, the high-resolution layout picture is transmitted back to cover the previous lower-resolution picture. By presetting the local high-definition function, the user is allowed to perform high-definition processing on specific areas according to actual needs, further improving the monitoring quality of key areas, reducing the requirements for network bandwidth, and meeting diversified and personalized needs. The partial panoramic picture transmission technology not only ensures the real-time performance of the data, but also effectively reduces the dependence on network bandwidth, so that the system can work stably in an environment with poor network conditions, meeting the application scenarios with high real-time requirements;

[0096] The low-light enhanced deep network carried by the present invention enables the system to clearly capture and identify targets at night or in dim light conditions, ensuring efficient operation around the clock and at all times. Combined with super-resolution deep network technology, the system can accurately identify targets at long distances and retain the fine details of the targets. Even if the targets are at the edge of the field of vision or at a farther position, high-quality recognition can be achieved. It shows significant advantages in functional diversity, all-weather adaptability, long-distance recognition capability, local high-definition function, recognition accuracy, panoramic field of view coverage, and real-time data transmission. It can better adapt to complex and changeable field conditions and complex task requirements, and provide users with more comprehensive, efficient, and personalized intelligent monitoring solutions.

[0097] Figure 2 It is a structural schematic diagram of the picture viewing device based on the smart pod provided by the present invention, and the picture viewing device based on the smart pod includes a first receiving unit 1, and the first receiving unit is used to receive a first viewing instruction of the user. The first viewing instruction is determined after the user selects a local picture perspective in the panoramic picture sent back by the smart pod. The panoramic picture is obtained by using the multi-channel cameras of the smart pod to obtain the original picture corresponding to each channel, and the first resolution picture is determined after all the original pictures are spliced. The working principle of the first receiving unit 1 can refer to the aforementioned step 101 and will not be repeated here.

[0098] The image viewing device based on the smart pod also includes a response unit 2, which is used to respond to the first viewing instruction, display the current display image in a preset display interface according to the local image viewing angle, determine the preloaded image according to the current display image and the preset magnification, and when it is determined that there is a target recognition object in the preloaded image, replace the first resolution corresponding target recognition object in the preloaded image with a second resolution corresponding target recognition object, the second resolution corresponding target recognition object is the enhanced image obtained after the smart pod performs resolution enhancement on the panoramic image, and is determined after transmitting back the position of the target recognition object in the enhanced image, the target recognition object is determined after the smart pod performs target recognition on the enhanced image using a preset target recognition model, the first resolution is smaller than the second resolution, and the working principle of the response unit 2 can refer to the aforementioned step 102, which will not be repeated here.

[0099] The image viewing device based on the smart pod also includes a second receiving unit 3, which is used to receive and respond to a second viewing instruction from the user, and the second viewing instruction is used to instruct the display of a target image with a target recognition object corresponding to the second resolution, determine the target image in the preloaded images, and display the target image in a preset display interface. The working principle of the second receiving unit 3 can refer to the aforementioned step 103 and will not be repeated here.

[0100] When performing panoramic picture preview, the present invention uses a lower-resolution picture to search for layout target recognition objects. When it is determined that the target recognition object exists, the high-resolution layout picture is transmitted back to cover the previous lower-resolution picture. By presetting the local high-definition function, the user is allowed to perform high-definition processing on specific areas according to actual needs, further improving the monitoring quality of key areas, reducing the requirements for network bandwidth, and meeting diversified and personalized needs. The partial panoramic picture transmission technology not only ensures the real-time performance of the data, but also effectively reduces the dependence on network bandwidth, so that the system can work stably in an environment with poor network conditions, meeting the application scenarios with high real-time requirements;

[0101] The low-light enhanced deep network carried by the present invention enables the system to clearly capture and identify targets at night or in dim light conditions, ensuring efficient operation around the clock and at all times. Combined with super-resolution deep network technology, the system can accurately identify targets at long distances and retain the fine details of the targets. Even if the targets are at the edge of the field of vision or at a farther position, high-quality recognition can be achieved. It shows significant advantages in functional diversity, all-weather adaptability, long-distance recognition capability, local high-definition function, recognition accuracy, panoramic field of view coverage, and real-time data transmission. It can better adapt to complex and changeable field conditions and complex task requirements, and provide users with more comprehensive, efficient, and personalized intelligent monitoring solutions.

[0102] Figure 3 Schematic diagram of the structure of the electronic device provided by the present invention. Figure 3 As shown, the electronic device may include: a processor 110, a communication interface 120, a memory 130 and a communication bus 140, wherein the processor 110, the communication interface 120 and the memory 130 communicate with each other through the communication bus 140. The processor 110 may call the logic instructions in the memory 130 to execute the image viewing method based on the smart pod, the method comprising: receiving a first viewing instruction from the user, the first viewing instruction being determined after the user selects a local image perspective in the panoramic image returned by the smart pod, the panoramic image being a first resolution image determined after the multi-channel cameras of the smart pod obtain the original images corresponding to each channel and all the original images are spliced; in response to the first viewing instruction, displaying the current display image in the preset display interface according to the local image perspective, determining the preloaded image according to the current display image and the preset magnification, and replacing the first in the preloaded image when it is determined that the target recognition object exists in the preloaded image. The target recognition object corresponding to the resolution is the target recognition object corresponding to the second resolution. The target recognition object corresponding to the second resolution is obtained after the smart pod performs resolution enhancement on the panoramic picture to obtain an enhanced picture. The enhanced picture is determined after the position of the target recognition object in the enhanced picture is transmitted back. The target recognition object is determined after the smart pod performs target recognition on the enhanced picture using a preset target recognition model. The first resolution is smaller than the second resolution. The second viewing instruction of the user is received and responded to. The second viewing instruction is used to instruct the display of a target picture with the target recognition object corresponding to the second resolution, the target picture is determined in the preloaded pictures, and the target picture is displayed in a preset display interface.

[0103] In addition, the logic instructions in the above-mentioned memory 130 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0104] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a picture viewing method based on a smart pod provided by the above methods, the method including: receiving a first viewing instruction from a user, the first viewing instruction is determined based on a local picture perspective selected by the user in a panoramic picture sent back by the smart pod, the panoramic picture is a first resolution picture determined by obtaining the original picture corresponding to each channel using multiple cameras of the smart pod and stitching all the original pictures; in response to the first viewing instruction, displaying the current display picture in a preset display interface according to the local picture perspective, determining a preloaded picture based on the current display picture and a preset magnification, and determining When there is a target recognition object in the preloaded image, the target recognition object corresponding to the first resolution is replaced with the target recognition object corresponding to the second resolution in the preloaded image, the target recognition object corresponding to the second resolution is obtained after the smart pod performs resolution enhancement on the panoramic image, and is determined after transmitting back the position of the target recognition object in the enhanced image, the target recognition object is determined after the smart pod performs target recognition on the enhanced image using a preset target recognition model, and the first resolution is smaller than the second resolution; receiving and responding to a second viewing instruction from the user, the second viewing instruction is used to instruct the display of a target image with the target recognition object corresponding to the second resolution, determining the target image in the preloaded image, and displaying the target image in a preset display interface.

[0105] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the image viewing method based on the smart pod provided by the above-mentioned methods, the method comprising: receiving a first viewing instruction from a user, the first viewing instruction being determined after a user selects a local image perspective in a panoramic image sent back by the smart pod, the panoramic image being a first resolution image determined after obtaining the original image corresponding to each channel using multiple cameras of the smart pod and stitching all the original images; in response to the first viewing instruction, displaying a currently displayed image in a preset display interface according to the local image perspective, determining a preloaded image according to the currently displayed image and a preset magnification, and determining whether a target recognition pair exists in the preloaded image. In the case of an image, the target recognition object corresponding to the first resolution is replaced with the target recognition object corresponding to the second resolution in the preloaded image, the target recognition object corresponding to the second resolution is obtained after the smart pod performs resolution enhancement on the panoramic image, and the enhanced image is obtained. The enhanced image is determined after the position of the target recognition object in the enhanced image is transmitted back, and the target recognition object is determined after the smart pod performs target recognition on the enhanced image using a preset target recognition model, and the first resolution is smaller than the second resolution; receiving and responding to a second viewing instruction from the user, the second viewing instruction is used to instruct the display of a target image with the target recognition object corresponding to the second resolution, determining the target image in the preloaded image, and displaying the target image in a preset display interface.

[0106] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0107] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for viewing pictures based on a smart pod, characterized in that: include: Receive a first viewing instruction from a user, where the first viewing instruction is determined based on a partial image perspective selected by the user in the panoramic image returned by the smart pod, where the panoramic image is a first resolution image determined by splicing all the original images after obtaining the original images corresponding to each channel using multiple cameras of the smart pod; In response to the first viewing instruction, a currently displayed image is displayed in a preset display interface according to the local image viewing angle, a preloaded image is determined according to the currently displayed image and a preset magnification, and when it is determined that a target recognition object exists in the preloaded image, a target recognition object corresponding to a first resolution is replaced in the preloaded image with a target recognition object corresponding to a second resolution, the target recognition object corresponding to the second resolution is obtained after the intelligent pod performs resolution enhancement on the panoramic image, and is determined after transmitting back the position of the target recognition object in the enhanced image, the target recognition object is determined after the intelligent pod performs target recognition on the enhanced image using a preset target recognition model, and the first resolution is smaller than the second resolution; Receive and respond to a second viewing instruction from the user, where the second viewing instruction is used to instruct display of a target image with a target recognition object corresponding to the second resolution, determine the target image from the preloaded images, and display the target image in a preset display interface.

2. The method for viewing pictures based on the smart pod according to claim 1, characterized in that: Before receiving the first viewing instruction from the user, the method further includes: The intelligent pod uses multiple cameras to collect images from all angles within the hemisphere to obtain the original images corresponding to each channel. The multiple cameras include a camera at the top of the hemisphere, four cameras corresponding to the middle of the hemisphere, and four cameras corresponding to the horizontal position of the hemisphere. According to the plane coordinates corresponding to each camera, each corresponding original image is mapped to the longitude and latitude coordinates corresponding to the hemisphere, and the hemispherical panoramic image of the first resolution is output.

3. The method for viewing pictures based on the smart pod according to claim 1, characterized in that: Before responding to the viewing instruction, the method further includes: For each original picture in the panoramic picture, when the brightness of any original picture is lower than a preset brightness, input the original picture into a low illumination enhancement network model to obtain an illumination enhanced picture output by the low illumination enhancement network model; Traversing all original pictures to determine all pictures to be processed, where the pictures to be processed include original pictures and / or illumination enhanced pictures; The low illumination enhancement network model is determined after training based on all low illumination sample pictures and illumination enhancement sample pictures corresponding to each low illumination sample picture.

4. The method for viewing pictures based on the smart pod according to claim 3, characterized in that: After all the pictures to be processed are determined, the method further includes: Input each image to be processed into a super-resolution enhancement network model, and obtain each resolution-enhanced image output by the super-resolution enhancement network model; splicing all the resolution-enhanced pictures to obtain the enhanced picture of the second resolution; The super-resolution enhancement network model is determined after training based on all sample processed images and the sample enhanced images corresponding to each sample processed image.

5. The method for viewing pictures based on the smart pod according to claim 4, characterized in that: After obtaining the enhanced image, the method further includes: Using a preset step length as a sliding speed, cropping the enhanced image using a preset slider window to obtain all cropped images; For each cropped image, input the cropped image to a preset target recognition model, and obtain a target recognition result output by the preset target recognition model, wherein the target recognition result includes the presence of the target recognition object and the absence of the target recognition object; The preset target recognition model is determined after training based on all sample recognition images containing target recognition objects and sample recognition images not containing target recognition objects.

6. The method for viewing pictures based on the smart pod according to claim 5, characterized in that: Before obtaining all the cropped images, the method further includes: According to the recognition type of the target recognition object, determining the window size value of the preset slider window from the corresponding relationship between the preset window size and the recognition type; According to the recognition type of the target recognition object, the step value of the preset step size is determined from the corresponding relationship between the preset step size and the recognition type.

7. The method for viewing pictures based on the smart pod according to claim 1, characterized in that: After determining the preloaded image according to the currently displayed image and the preset magnification, the method further includes: When it is determined that all target recognition objects exist in the preloaded image, replacing the target recognition objects corresponding to the first resolution with the target recognition objects corresponding to the second resolution in the preloaded image; When it is determined that some target recognition objects exist in the preloaded image, the preset magnification is increased, a preloaded image corresponding to all target recognition objects is determined, and the target recognition objects corresponding to the first resolution are replaced with the target recognition objects corresponding to the second resolution in the preloaded image.

8. The method for viewing pictures based on the smart pod according to claim 1, characterized in that: After receiving and responding to the second viewing instruction from the user, the method further includes: At every preset time interval, the updated panoramic picture returned by the smart pod is updated in the preset display interface, and again in response to the first viewing instruction, the updated display picture is displayed in the preset display interface according to the local picture perspective, and again the second viewing instruction of the user is received and responded to, and the updated picture with the target recognition object corresponding to the second resolution is displayed.

9. A picture viewing device based on a smart pod, characterized in that: include: A first receiving unit, the first receiving unit is used to receive a first viewing instruction of a user, the first viewing instruction is determined after the user selects a local image perspective in the panoramic image returned by the smart pod, the panoramic image is a first resolution image determined after the original image corresponding to each channel is obtained by using multiple cameras of the smart pod and all the original images are stitched together; a response unit, the response unit being used to respond to the first viewing instruction, display the current display image in a preset display interface according to the local image viewing angle, determine the preloaded image according to the current display image and the preset magnification, and replace the first resolution corresponding target recognition object with the second resolution corresponding target recognition object in the preloaded image when it is determined that there is a target recognition object in the preloaded image, the second resolution corresponding target recognition object is obtained after the intelligent pod performs resolution enhancement on the panoramic image, and is determined after transmitting back according to the position of the target recognition object in the enhanced image, the target recognition object is determined after the intelligent pod performs target recognition on the enhanced image using a preset target recognition model, and the first resolution is smaller than the second resolution; A second receiving unit, the second receiving unit is used to receive and respond to a second viewing instruction of the user, the second viewing instruction is used to instruct the display of a target image with a target recognition object corresponding to the second resolution, determine the target image in the preloaded images, and display the target image in a preset display interface.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the image viewing method based on the smart pod as described in any one of claims 1 to 8 is implemented.

Images