Rapid target tracking shooting system based on multiple prime lenses

Abstract

The invention relates to the technical field of optical imaging and image processing, and particularly discloses a fast target tracking shooting system based on multiple prime lenses, which comprises a plurality of independent prime lenses physically arranged side by side, a high-speed image processing unit, a precise holder and a central control unit. Through parallel processing of multiple paths of video streams, target tracking and state prediction are carried out under a unified global coordinate system, and based on a prediction result, a preferred lens is directly switched and a pre-stored focusing parameter is called, so that instantaneous and unfuzzy switching and full-time clear imaging of a field of view and a focal length are realized; and the response speed, the stability and the imaging quality of high-speed target tracking are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of optical imaging and image processing technology, specifically relating to a fast target tracking and shooting system based on multiple fixed-focus lenses. Background Technology

[0002] In the field of visual information acquisition and processing, target tracking and imaging technology is a key means to achieve continuous observation and high-quality image recording of dynamic targets. It is widely used in many important scenarios such as security monitoring, wildlife observation, sports event broadcasting, and military reconnaissance. The core of this technology lies in how to quickly and accurately lock onto moving targets from complex backgrounds and acquire clear and stable image data in real time. Its performance directly determines the effectiveness of subsequent analysis and applications.

[0003] Active tracking shooting systems based on gimbals and optical lenses represent the mainstream technology for achieving the aforementioned functions. These systems typically use a gimbal to rotate in azimuth and pitch to follow the target, and rely on a zoom lens to adjust the field of view to obtain target images at different scales. The basic working logic is as follows: first, the wide-angle end of the lens is used to acquire a large field of view for initial target screening and coarse tracking; then, the lens is zoomed to the telephoto end to narrow the field of view and magnify the target, thereby capturing close-up images with richer details.

[0004] The existing gimbal + single zoom lens architecture has inherent performance bottlenecks. The zoom lens requires a long mechanical drive time during focus switching, causing the system to fail to acquire a clear target image during zooming. Simultaneously, to adapt to changes in target distance, the lens needs to frequently perform autofocus operations, a process that is also time-consuming and may lose focus or fail to focus due to rapid target movement. These two consecutive mechanical delays severely restrict the overall response speed of the system, making it highly susceptible to losing the target or only acquiring blurry images when tracking high-speed moving targets due to zoom and focus lag. Furthermore, frequent mechanical movements not only reduce system reliability but also increase power consumption and wear. Therefore, fundamentally eliminating the mechanical delays of zoom and focus to achieve near-instantaneous field-of-view switching and always-clear imaging has become a crucial technical challenge to overcome in improving the performance of fast target tracking and shooting systems. Summary of the Invention

[0005] The purpose of this invention is to provide a fast target tracking and shooting system based on multiple fixed-focus lenses, so as to solve the problems of slow system response, easy loss of tracking high-speed targets, and unstable image quality caused by mechanical delays in the existing technology due to reliance on a single zoom lens for focal length switching and focusing.

[0006] This invention provides a rapid target tracking and shooting system based on multiple fixed-focus lenses. This system abandons the traditional single-zoom lens architecture and instead employs a collaborative working system consisting of multiple independent fixed-focus lenses, a high-speed image processing unit, a precision gimbal, and a central control unit. The core innovation of this system lies in the physical juxtaposition of multiple fixed-focus lenses with different focal lengths, pre-calibrating the focus of each lens, and using the high-speed image processing unit to perform real-time parallel analysis and fusion of multiple video streams. This completely eliminates the mechanical delay of zooming and autofocus, achieving seamless, instantaneous switching from wide-area search to close-up tracking and all-time clear imaging.

[0007] The system includes:

[0008] A high-speed image processing unit is used to receive and process real-time video streams from all fixed-focus lenses in parallel, performing target detection, multi-scale feature matching, and tracking trajectory prediction.

[0009] The lens array module consists of at least three independent optical fixed-focus lenses physically juxtaposed, each with a different fixed focal length, covering a continuous or discrete field of view from wide-angle to telephoto. The optical axes of all fixed-focus lenses are pre-aligned mechanically and optically with the imaging sensor plane, and the focus calibration is pre-performed on multiple typical object distance planes.

[0010] A precision gimbal is used to support and drive the entire lens array module to rotate in azimuth and pitch.

[0011] The central control unit coordinates the operation of the high-speed image processing unit and the precision gimbal. Based on the target status information output by the high-speed image processing unit, it generates gimbal control commands and decides the preferred shot for outputting the main video stream.

[0012] Furthermore, the lens array module is specifically configured as follows: The module includes a wide-angle fixed-focus lens, a medium-range fixed-focus lens, and a telephoto fixed-focus lens. The wide-angle fixed-focus lens has a horizontal field of view greater than 60 degrees, used for wide-area scene monitoring and initial target acquisition. The medium-range fixed-focus lens has a horizontal field of view between 20 and 40 degrees, used for stable tracking of acquired targets and observation of medium-level details. The telephoto fixed-focus lens has a horizontal field of view less than 10 degrees, used for high-resolution close-up shooting of targets. The projection position deviation of the optical center of all fixed-focus lenses in a plane perpendicular to the optical axis is less than 0.1 mm, and the imaging sensors of each lens have the same photoelectric response characteristics and resolution.

[0013] Furthermore, the internal processing flow of the high-speed image processing unit is as follows: This unit includes multiple video input interfaces, a parallel computing core, and a fusion output interface. The multiple video input interfaces synchronously latch the raw image data output from all fixed-focus lenses. The parallel computing core allocates an independent processing thread to each video stream. Each thread executes the same target detection algorithm, but uses detection parameters that match the field-of-view characteristics of the corresponding lens for that video stream. Specifically, for wide-angle lens video streams, the detection algorithm uses a lower detection confidence threshold and a larger candidate box size to prioritize high recall; for telephoto lens video streams, the detection algorithm uses a higher detection confidence threshold and a finer candidate box size to improve detection accuracy and suppress background interference. The target information detected by all threads, including the target's position, size, and feature descriptors in their respective image coordinate systems, is uniformly sent to the global association and tracking module.

[0014] Furthermore, the global association and tracking module operates as follows: First, the module establishes a unified global world coordinate system, with its origin located at the gimbal rotation center. The module receives target detection results from different camera threads and, using pre-calibrated pose parameters of each camera relative to the gimbal rotation center, maps the two-dimensional position of each target in its image coordinate system, combined with the focal length and focus distance parameters of the camera, to the unified global world coordinate system through perspective projection model inverse calculation, obtaining a three-dimensional position estimate. Next, the module associates and matches the three-dimensional target points calculated by all cameras in the current frame with the target tracking trajectories established in the previous frame. The matching algorithm comprehensively calculates based on the target's three-dimensional position, motion velocity vector, and cross-camera appearance feature similarity. For successfully associated targets, the module updates their motion state using a Kalman filter and predicts their position in the global coordinate system at the next moment.

[0015] Furthermore, the decision-making logic of the central control unit is as follows: The central control unit receives in real time the status information of all tracked targets in the global coordinate system, including position, velocity, and the tracking identifier, output by the global correlation and tracking module. When the system is in wide-area search mode, the central control unit designates the video stream of the wide-angle fixed-focus lens as the main output stream of the system. Once any target is continuously and stably tracked for more than 5 frames, the central control unit initiates a lens switching decision. The decision is based on the expected imaging size of the target in the image. The central control unit calculates the pixel size occupied by the target in the corresponding image if it is imaged by different lenses, based on the predicted position of the target in the global coordinate system, the system's preset ideal target imaging size, and the focal length parameters of each fixed-focus lens. The central control unit selects the fixed-focus lens that makes the target imaging size closest to the preset ideal value, designates it as the new main output lens, and simultaneously sends a control command to the precision gimbal to rotate the gimbal, ensuring that the optical axis of the preferred lens is aligned with the predicted position of the target.

[0016] Furthermore, the control method of the precision gimbal is as follows: The precision gimbal receives the predicted azimuth and pitch angles of the target in the global coordinate system sent by the central control unit. The high-precision encoder built into the gimbal provides real-time feedback on the current angle of the gimbal. The gimbal controller uses a proportional-integral-derivative (PID) control algorithm to calculate the deviation between the current angle and the target angle, and generates corresponding motor drive signals to enable the gimbal to smoothly and rapidly point the lens array module towards the target direction. The upper limits of the gimbal's angular velocity and angular acceleration are set according to the requirements of mechanical structural strength and imaging stability, with a maximum angular velocity of not less than 60 degrees per second.

[0017] Furthermore, the focusing guarantee mechanism of the system is as follows: Since each fixed-focus lens has been pre-calibrated on multiple object distance planes and its corresponding focusing motor position parameters are stored, during actual operation, the central control unit queries the nearest focusing calibration plane based on the target's distance information in the global coordinate system and directly sends the corresponding focusing motor position parameters to the lens's focusing driver. This process is open-loop control and does not rely on real-time image contrast feedback; therefore, the focusing action can be completed within 1 millisecond, and there is no search process like in traditional autofocus, thus ensuring that the selected fixed-focus lens can immediately output a clear image at any switching moment.

[0018] Furthermore, the system also includes a multi-video stream fusion and output module. This module receives real-time video streams from all fixed-focus lenses and a main lens identifier designated by the central control unit. The module outputs the main lens's video stream as the full-resolution main screen. Simultaneously, the module selects at least one video stream from non-main lenses, scales it down proportionally, and overlays it onto the main screen in a picture-in-picture or edge-split-screen format to assist in displaying broader background information or other areas of interest. All video streams are tagged with a unified timestamp and target tracking metadata.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] 1. This invention, by employing a physically juxtaposed architecture of multiple fixed-focus lenses, fundamentally eliminates the mechanical drive time required for focal length switching in a single zoom lens. The system electronically selects the output video stream from lenses with different focal lengths, with a switching delay of only microseconds—a significant improvement in field-of-view switching speed compared to the hundreds of milliseconds of delay in traditional optical zoom. This allows the system to instantly respond to scale changes in high-speed moving targets, greatly reducing the probability of losing targets due to zoom lag.

[0021] 2. This invention pre-calibrates the focus of each fixed-focus lens on multiple object distance planes and transforms the focus control from a traditional closed-loop search mode based on image feedback to an open-loop preset mode based on target distance query. This completely avoids the lens repetitive search time and potential focus failure problems during autofocus. Regardless of changes in target distance, the system can directly call the pre-stored focus parameters to complete focusing in a very short time, ensuring that the selected lens can output a clear image at all times, achieving all-time clear imaging and significantly improving the stability and reliability of image quality.

[0022] 3. This invention utilizes a high-speed image processing unit to process multiple video streams in parallel and correlate them with global coordinates, constructing a collaborative perception network that transcends the limitations of a single lens's field of view. The system comprehensively leverages the wide field of view of a wide-angle lens for target detection and the high resolution of a telephoto lens for detail capture, maintaining continuous and stable target tracking in the global coordinate system through a unified tracking algorithm. This multi-scale collaborative mechanism not only enhances the system's robustness in complex scenes but also achieves smooth and precise tracking through predictive gimbal control and lens selection, resulting in a high-performance target tracking and shooting solution that is responsive, produces clear images, and provides stable tracking. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall technical solution architecture of the fast target tracking and shooting system based on multiple fixed-focus lenses proposed in this invention;

[0024] Figure 2 This is a schematic diagram of the core principle framework of multi-fixed-focus lens collaborative sensing and instantaneous switching in this invention;

[0025] Figure 3 This is a logical flowchart of the high-speed image processing unit performing parallel processing of multiple video streams and global correlation in this invention.

[0026] Figure 4 This is a flowchart illustrating the logical flow of the central control unit's lens selection decision and gimbal control coordination in this invention.

[0027] Figure 5 This is a schematic diagram of the multi-level interaction and data flow of the system in this invention, from wide-area search to close-up tracking. Detailed Implementation

[0028] Example 1: The overall architecture and operating logic of the fast target tracking and shooting system based on multiple fixed-focus lenses proposed in this invention can be combined with the appendix. Figure 1To gain a complete understanding, this system abandons the traditional mechanical focus adjustment mechanism that relies on a single zoom lens. Instead, it constructs a collaborative working system consisting of multiple physically juxtaposed independent fixed-focus lenses, a high-speed image processing unit, a precision gimbal, and a central control unit. The components are highly coupled in terms of spatial layout, data interaction, and control timing, working together to achieve seamless switching from wide-area search to high-resolution close-up tracking, and ensuring that the output video stream is always in sharp focus at any given time.

[0029] Please refer to the attached document. Figure 1 The core hardware modules of the system include a lens array module, a high-speed image processing unit, a precision gimbal, and a central control unit. The lens array module consists of three fixed-focus lenses: a wide-angle fixed-focus lens, a medium-focus fixed-focus lens, and a telephoto fixed-focus lens. These three lenses are physically mounted side-by-side along the same plane, with their optical axes parallel to each other, and the imaging sensor planes of all lenses are strictly coplanar. To ensure consistency in multi-lens imaging and accuracy in coordinate mapping, the projection position deviation of the optical center of each lens in the direction perpendicular to the optical axis is controlled within 0.1 mm. Furthermore, all imaging sensors use the same type of complementary metal-oxide-semiconductor (CMOS) image sensor, possessing consistent pixel size, dynamic range, signal-to-noise ratio, and frame rate response characteristics, with a typical frame rate of 30 or 60 frames per second and a resolution of no less than 1920×1080 pixels.

[0030] The wide-angle prime lens has a horizontal field of view of 75 degrees, designed to cover a large monitoring area and suitable for the initial target acquisition phase. The mid-range prime lens has a horizontal field of view of 30 degrees, providing stable tracking and medium-detail imaging once the target enters the mid-range. The telephoto prime lens has a horizontal field of view of 8 degrees, specifically designed for high-resolution close-up shooting of close-up or high-value targets. Together, these three lenses form a discrete yet continuous focal length sequence covering everything from wide-field to close-up, effectively avoiding the delay and shake caused by mechanical movement during focal length switching in traditional zoom lenses.

[0031] All fixed-focus lenses undergo multi-object distance plane focusing calibration before leaving the factory. Specifically, at six typical object distance planes (1m, 3m, 5m, 10m, 20m, and 50m) from the lens, the focusing motor is adjusted to the optimal imaging position, and the corresponding motor drive voltage value or stepper motor position code is stored in non-volatile memory. This calibration process is jointly completed by a high-precision laser rangefinder and an autofocus evaluation function, ensuring that the actual image sharpness at each stored position reaches or exceeds a preset threshold. During system operation, once a fixed-focus lens is selected as the main output lens, the central control unit queries the nearest calibrated object distance plane based on the real-time distance information of the target in the global coordinate system and directly sends the corresponding motor position parameters to the focusing driver of that lens. This focusing process is open-loop control and does not rely on image contrast feedback, thus it can be completed within 1 millisecond, completely avoiding the time loss and focus failure risk caused by repeated searches in traditional autofocus algorithms.

[0032] The high-speed image processing unit synchronously receives raw image data streams from three fixed-focus lenses via a multi-channel video input interface. This interface employs a hardware-level frame synchronization mechanism to ensure strict alignment of the three video streams in terms of timestamps, with a maximum time deviation of no more than 1 microsecond. The received data is fed into a parallel computing core, which consists of multiple graphics processing units or dedicated vision processing chips, supporting multi-threaded concurrent execution. Each video stream is assigned an independent processing thread, and each thread runs the same object detection algorithm in parallel—such as a deep learning-based YOLOv5 or Faster R-CNN model—but with differentiated detection parameters configured for the field of view characteristics of different lenses.

[0033] For video streams from wide-angle fixed-focus lenses, the confidence threshold for the detection algorithm is set to 0.3, and the maximum size of the candidate box is limited to 0.8 times the image height to prioritize high recall and prevent missing small targets in large scenes. For mid-range fixed-focus lenses, the confidence threshold is set to 0.5, and the candidate box size is limited to 0.5 times the image height. For telephoto fixed-focus lenses, the confidence threshold is increased to 0.7, and the candidate box size is limited to 0.3 times the image height, thereby suppressing background noise interference in high-resolution images and improving detection accuracy. The target information output by each thread includes: target category, bounding box coordinates (with the top left corner of the image as the origin), target center pixel coordinates, target scale (represented by pixel area), and a 128-dimensional feature descriptor extracted by the convolutional neural network.

[0034] The aforementioned multi-path detection results are uniformly sent to the global correlation and tracking module. This module first establishes a global world coordinate system with the gimbal rotation center as the origin, with its X-axis pointing due north, Y-axis pointing to the zenith, and Z-axis forming a right-handed coordinate system. Using pre-calibrated rigid body pose parameters (including translation vectors and rotation matrices) of each lens relative to the gimbal rotation center, combined with the current real-time azimuth and pitch angle readings of the gimbal, the module maps the two-dimensional pixel position of each target in its respective image coordinate system to a three-dimensional point in the global world coordinate system through perspective projection inverse calculation.

[0035] This mapping process follows the standard pinhole camera model. Let the pixel coordinates of a target in the i-th lens image be... The focal length of this lens is (Unit: pixels), principal point coordinates are The extrinsic parameter of the lens relative to the center of rotation of the gimbal is the rotation matrix. With translation vector The rotation matrix corresponding to the current attitude of the gimbal is: The target's three-dimensional position in the global coordinate system. It can be calculated using the following formula:

[0036]

[0037] in, This is the depth scale factor, determined by the target's focusing distance (i.e., the calibration object distance). Since the system knows the object distance plane where the target is located, The depth value corresponding to the object distance can be directly obtained without the need for stereo matching or deep learning estimation, thereby significantly improving the real-time performance and determinism of 3D positioning.

[0038] After obtaining the 3D target points calculated from all shots in the current frame, the global association and tracking module associates them with the tracking trajectories established in the previous frame. Each tracking trajectory includes the target's historical position sequence, velocity vector, acceleration estimate, and a unique tracking identifier. The association process employs a multi-hypothesis fusion strategy, comprehensively considering the similarity in the following three dimensions: first, the distance between 3D spatial positions, using Euclidean distance as a metric, with a threshold of 2 meters; second, motion consistency, i.e., the angle between the velocity vectors of the current observation point and the trajectory prediction point is less than 30 degrees; and third, cross-shot appearance feature similarity, calculated by the cosine similarity between the current target feature descriptor and the historical average features of the trajectory, with a threshold of 0.6. If an observation point and a trajectory meet all three criteria, the association is considered successful.

[0039] For a successfully associated target, the module initiates a Kalman filter to update its motion state. The state vector is defined as follows: , representing the target's three-dimensional position and velocity in the global coordinate system, respectively. The process noise covariance matrix is ​​dynamically adjusted according to the target's motion type (e.g., pedestrian, vehicle, aircraft), while the measurement noise covariance is weighted based on the resolution and signal-to-noise ratio of the current lens. The filter outputs an updated state estimate and predicts the target's three-dimensional position in the next frame based on a uniform motion model.

[0040] The central control unit continuously receives status information of all tracked targets from the global correlation and tracking module. The system initially operates in wide-area search mode, where the central control unit designates the video stream from the wide-angle fixed-focus lens as the main output stream. When a target is continuously and stably tracked for more than 5 frames, and its predicted position is within the effective field of view of a medium- or telephoto lens, the central control unit initiates the lens selection decision process.

[0041] This decision is based on the target's expected imaging size in the image. The system's preset ideal target imaging height is 0.4 times the image height. The central control unit determines the target's position based on its predicted location in the global coordinate system. Focal length of each prime lens and the pixel size of the imaging sensor Calculate the pixel height of the target in the image when it is imaged by the i-th lens. The calculation formula is as follows:

[0042]

[0043] in, The actual height of the target (which can be preset by prior knowledge of the category, such as 1.7 meters for pedestrians and 1.5 meters for vehicles). The straight-line distance from the optical center of the lens to the target is given by... Calculations show that Let be the optical center position of the i-th lens in the global coordinate system. The central control unit iterates through all lenses and selects the one that... The smallest lens is selected as the preferred lens, among which The image height (e.g., 1080 pixels).

[0044] Once the preferred lens is determined, the central control unit immediately performs two operations: first, it sends a main lens identifier to the multi-stream fusion and output module, instructing it to switch the main image source; second, it sends the predicted azimuth angle θ and pitch angle ϕ of the target in the global coordinate system to the precision gimbal. The azimuth angle θ is determined by... Calculate the pitch angle ϕ from... Calculation, where Predict the location vector for the target.

[0045] After receiving the angle command, the precision gimbal initiates closed-loop servo control. The gimbal's built-in absolute photoelectric encoder provides feedback on the current azimuth and pitch angles at a frequency of 1000 times per second. The gimbal controller employs a proportional-integral-derivative (PID) control algorithm, with the following control law:

[0046]

[0047] in, Due to angular deviation, , , The control gain is determined by the system. To balance response speed and imaging stability, the maximum angular velocity of the gimbal is limited to 60 degrees per second, and the maximum angular acceleration is 300 degrees squared per second. In typical tracking scenarios, the gimbal can complete precise pointing from wide-angle to telephoto fields of view within 0.5 seconds, with smooth and jitter-free movement, ensuring no significant shaking in the video stream.

[0048] The multi-stream video fusion and output module receives three raw video streams and a main camera identifier. This module outputs the main camera video stream at full resolution as the main screen. Simultaneously, it selects one video stream from the non-main camera (usually a wide-angle lens), scales it to 20% of the main screen width, and overlays it in a picture-in-picture format in the lower right corner of the main screen. The position, transparency, and border style of the overlay area are configurable. All output video frames are embedded with a unified timestamp (with microsecond-level precision) and target tracking metadata, including tracking identifier, target category, 3D position, velocity vector, and current main camera number, facilitating subsequent intelligent analysis or manual verification.

[0049] The entire system's workflow can be combined with the appendix Figure 5 Further explanation: In the initial stage, the wide-angle lens continuously scans a large area, while the high-speed image processing unit runs a low-threshold detection algorithm in parallel. Once a potential target is detected, the global correlation module attempts to establish a tracking trajectory. If the trajectory continuity meets the requirements, the central control unit assesses the target scale and decides whether to switch to the medium or telephoto lens. After the switch command is issued, the focus driver immediately loads pre-stored parameters, and the gimbal synchronously rotates to point at the predicted target position. During this process, the high-speed image processing unit continuously processes three video streams to ensure that the system can maintain target perception and tracking even at the moment of switching. After the switch is completed, the main screen seamlessly transitions to the new lens, and the image clarity is not diminished in any way.

[0050] This system achieves a qualitative leap in target tracking and shooting performance through triple innovations in hardware parallelization, open-loop control, and global perception. In actual testing, the system can switch lenses and point the gimbal 0.3 seconds before a vehicle target moving laterally across the field of view at a speed of 20 meters per second enters the telephoto field of view, with no target loss throughout the process. Moreover, the modulation transfer function values ​​of all output frames are higher than 0.3, meeting the requirements for high-definition imaging.

[0051] Example 2: Based on the previous examples, this example introduces a dynamic focal length interpolation mechanism to address scenarios where there are gaps in focal length coverage within the lens array. Specifically, when the image size of the target in the image is between the ideal imaging range of two adjacent fixed-focus lenses, the system does not directly switch to either physical lens, but instead activates a virtual intermediate focal length synthesis mode.

[0052] This mode is executed by the super-resolution reconstruction submodule within the high-speed image processing unit. This submodule receives synchronized video frames from two adjacent physical lenses (e.g., mid-range and telephoto) and estimates the ideal imaging shape of the target at a virtual intermediate focal length using optical flow based on the target's position, scale, and subpixel displacement information in the two frames. Subsequently, a generative adversarial network based on an attention mechanism is used to fuse the contextual information of the mid-range image with the high-frequency details of the telephoto image, generating a synthetic image equivalent to an intermediate focal length (e.g., 45mm). The resolution of this synthetic image is consistent with the original sensor, and its edge sharpness is subjectively evaluated to be superior to that of linear interpolation.

[0053] When making decisions, the central control unit considers such virtual lenses as one of the viable options. If calculations show that a virtual lens can more accurately match the preset imaging size, it is designated as the logical master lens, and the composite image is output as the main view. At this time, the physical gimbal still points to the actual target location, but the focus parameters are taken from the physical lens closest to the target. This mechanism effectively extends the system's effective focal length continuity and improves its adaptability to non-standard target scales.

[0054] Furthermore, this embodiment enhances the global association and tracking module. When a target is occluded or briefly lost, the module no longer immediately terminates the trajectory but instead initiates a trajectory continuation mechanism based on kinematic constraints. This mechanism utilizes the target's historical velocity and acceleration, combined with a scene semantic map (such as road directions and building boundaries), to predict the target's possible reappearance area. Within this area, the detection confidence threshold is lowered, extending the trajectory survival time to 10 frames. If the target reappears within the predicted area, the trajectory automatically resumes, avoiding frequent trajectory reconstruction and label switching, thus improving the continuity of the user experience.

[0055] The control strategy for the precision gimbal has also been optimized. In high-speed tracking mode, the gimbal controller introduces a feedforward compensation term. This term, based on the angular velocity and angular acceleration predicted by the Kalman filter, applies control torque in advance to counteract the phase lag caused by mechanical inertia. Experimental results show that this strategy can reduce the gimbal pointing error by 40%, with particularly significant effects when the target is making sharp turns.

[0056] All the aforementioned enhancement mechanisms are implemented through software algorithm upgrades without altering the core hardware architecture, demonstrating the excellent scalability and engineering practicality of this invention. The system exhibits superior robustness and real-time performance in various scenarios, including complex urban roads, sporting events, and wildlife observation, fully validating the effectiveness and advancement of the technical solution.

Claims

1. A fast target tracking and shooting system based on multiple fixed-focus lenses, characterized in that, include: The lens array module consists of at least three independent optical fixed-focus lenses physically juxtaposed, each with a different fixed focal length, covering a field of view from wide-angle to telephoto. The optical axes of all fixed-focus lenses are pre-aligned mechanically and optically with the imaging sensor plane, and the focus calibration is pre-performed on multiple typical object distance planes. A high-speed image processing unit is used to receive and process real-time video streams from all fixed-focus lenses in parallel, performing target detection, multi-scale feature matching, and tracking trajectory prediction. A precision gimbal is used to support and drive the entire lens array module to rotate in azimuth and pitch. The central control unit coordinates the operation of the high-speed image processing unit and the precision gimbal. Based on the target status information output by the high-speed image processing unit, it generates gimbal control commands and decides the preferred shot for outputting the main video stream.

2. The fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 1, characterized in that, The high-speed image processing unit includes a global association and tracking module, which is used to establish a unified global world coordinate system with the gimbal rotation center as the origin, receive target detection results from different fixed-focus lenses, and use the pre-calibrated pose parameters of each lens relative to the gimbal rotation center, combined with the focal length and focusing distance parameters of the lens, to perform inverse calculation through perspective projection model, and map the two-dimensional position of each target in its image coordinate system to the unified global world coordinate system to obtain a three-dimensional position estimate. The three-dimensional target points calculated from all shots in the current frame are associated and matched with the target tracking trajectories established in the previous frame. For targets that are successfully associated, their motion state is updated using a Kalman filter, and their position in the global coordinate system at the next moment is predicted.

3. The fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 2, characterized in that, The central control unit receives in real time the status information of all tracked targets in the global coordinate system output by the global correlation and tracking module; when the system is in wide-area search mode, the video stream of the wide-angle fixed-focus lens is designated as the main output stream of the system. When any target is continuously and stably tracked for more than a preset number of frames, a lens switching decision is initiated. The decision is based on the expected imaging size of the target in the image. The central control unit calculates the pixel size of the target in the corresponding image if it is imaged by different lenses, based on the predicted position of the target in the global coordinate system, the ideal target imaging size preset by the system, and the focal length parameters of each fixed-focus lens. Select a fixed-focus lens that can make the target image size closest to the preset ideal value, designate it as the new main output lens, and at the same time send control commands to the precision gimbal to rotate the gimbal to ensure that the optical axis of the preferred lens is aligned with the predicted position of the target.

4. The fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 3, characterized in that, The lens array module includes a wide-angle fixed-focus lens, a medium-focus fixed-focus lens, and a telephoto fixed-focus lens; The wide-angle fixed-focus lens is used for large-area scene monitoring and initial target acquisition; The medium-focus fixed-focus lens is used for stable tracking and medium-detail observation of the captured target; The telephoto fixed-focus lens is used for high-resolution close-up shooting of the target.

5. A fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 4, characterized in that, The high-speed image processing unit also includes a multi-channel video input interface and a parallel computing core; The multi-channel video input interface is used to synchronously latch the raw image data output by all fixed-focus lenses; The parallel computing core allocates an independent processing thread to each video stream. Each thread executes the same target detection algorithm, but uses detection parameters that match the field of view characteristics of the corresponding lens in that video stream. For wide-angle video streams, the detection algorithm uses a lower detection confidence threshold and a larger candidate box size; For telephoto lens video streams, the detection algorithm employs a high detection confidence threshold and fine candidate box size.

6. A fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 5, characterized in that, The association matching algorithm executed by the global association and tracking module performs a comprehensive calculation based on the target's three-dimensional position, motion velocity vector, and cross-shot appearance feature similarity.

7. A fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 6, characterized in that, The precision gimbal receives the predicted azimuth and elevation angles of the target in the global coordinate system sent by the central control unit. The gimbal's built-in high-precision encoder provides real-time feedback on the gimbal's current angle. The gimbal controller uses a proportional-integral-derivative (PID) control algorithm to calculate the deviation between the current angle and the target angle, and generates corresponding motor drive signals to enable the gimbal to smoothly and quickly point the lens array module toward the target direction.

8. A fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 7, characterized in that, The system also includes a focus protection mechanism; Each fixed-focus lens is pre-calibrated on multiple object distance planes and the corresponding lens focusing motor position parameters are stored. In actual operation, the central control unit queries the nearest focusing calibration plane based on the distance information of the target in the global coordinate system, and sends the corresponding focusing motor position parameters directly to the focusing driver of the lens. The focusing process is an open-loop control that does not rely on real-time image contrast feedback.

9. A fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 8, characterized in that, The system also includes a multi-video stream fusion and output module; This module receives real-time video streams from all fixed-focus lenses and the main lens identifier specified by the central control unit; The module outputs the main camera's video stream as the full-resolution main screen. At the same time, the module selects at least one video stream from non-main camera, scales it down proportionally, and overlays it on the main screen in the form of picture-in-picture or edge split screen. All output video streams are embedded with a unified timestamp and target tracking identifier metadata.

10. A fast target tracking and shooting system based on multiple fixed-focus lenses according to claim 9, characterized in that, The process by which the central control unit calculates the target imaging size is as follows: Based on the target's predicted position in the global coordinate system, the target's actual height, the focal length of each fixed-focus lens, and the pixel size of the imaging sensor, calculate the pixel height of the target in the image if it is imaged by the i-th lens.

Images