Search

CN-122027907-A - Night vision rifle bolt of multilane snap shot

CN122027907ACN 122027907 ACN122027907 ACN 122027907ACN-122027907-A

Abstract

The invention discloses a multi-lane snapshot night vision gun camera, which relates to the field of video monitoring equipment and comprises a multi-lane synchronous imaging module, an algorithm pre-processing unit, an AI scene segmentation and identification system, a night vision full-color optimization module and a snapshot cooperative control unit. The multi-lane synchronous imaging module generates independent lane image streams by using a multi-sensor array to avoid target interference, the algorithm pre-processing unit is used for preprocessing images in a lightweight mode, bandwidth extraction efficiency is reduced, the AI system is used for dividing and identifying vehicles and lane lines in a multi-branch mode, associated tracks are used for generating snapshot conditions, the night vision module is used for dynamically adjusting parameters according to AI feedback and realizing full-color imaging without light supplement, and the snapshot unit is used for synchronizing shutters and marking lane attributes. The invention solves the problems of overlapping imaging, night vision color difference and snapshot delay of the traditional gun camera, improves the accuracy and efficiency of multi-lane snapshot at night, and adapts to complex traffic scenes.

Inventors

  • Ye Shuiqian

Assignees

  • 珠海互通微电子有限公司

Dates

Publication Date
20260512
Application Date
20260209

Claims (10)

  1. 1. A multi-lane snapshot night-vision gun camera is used for video monitoring processing and is characterized by comprising a multi-lane synchronous imaging module, an algorithm pre-processing unit, an AI scene segmentation recognition system, a night-vision full-color optimization module and a snapshot cooperative control unit, wherein the multi-lane synchronous imaging module adopts a multi-sensor array architecture, each sensor corresponds to an independent lane imaging area, can synchronously acquire night image data of each lane and generate lane independent image streams to avoid target interference caused by overlapping of multi-lane images, the algorithm pre-processing unit is integrated between the imaging module and an AI system, a lightweight image preprocessing algorithm is built in, real-time noise reduction, edge enhancement and dynamic exposure correction are carried out on each lane independent image stream, data transmission quantity is reduced, and subsequent AI processing efficiency is improved, the AI scene segmentation recognition system receives the lane image data subjected to pre-processing, carries out scene semantic segmentation on each lane to distinguish vehicle lane line by a multi-branch segmentation network, and simultaneously calls a target tracking sub-model to realize cross-frame vehicle track association, and generates a precise snapshot condition in combination with lane independent image streams, the night-vision preprocessing algorithm is integrated between the imaging module and the AI system, and the full-color image capturing system is in real-time, and the full-color image capturing algorithm is not required to be matched with the full-color image capturing device, and the full-color image capturing system is in real-time, and the full-color image capturing image is not required to be controlled to be in real-time, and the full-color image capturing condition is not to be matched with the full-color image capturing system, and has a real-time image capturing condition, and has a real-time image capturing image conditions.
  2. 2. The multi-lane snapshot night vision gun camera according to claim 1, wherein the multi-sensor array of the multi-lane synchronous imaging module is arranged in a coaxial line array, each sensor is provided with an independent optical lens and a vision angle adjusting component, the imaging areas of the sensors can completely cover the corresponding lanes and are not overlapped through precise matching of the lens focal length and the vision angle, and meanwhile, a time synchronization circuit is built in the module, so that all the sensors are ensured to trigger exposure at the same time, and the cross-lane vehicle snapshot dislocation caused by exposure time difference is avoided.
  3. 3. The multi-lane snapshot night vision gun camera according to claim 2, wherein the lightweight image preprocessing algorithm of the algorithm preprocessing unit comprises a lane self-adaptive noise reduction sub-algorithm and a dynamic exposure correction sub-algorithm, wherein the lane self-adaptive noise reduction sub-algorithm selects a filtering strategy according to noise characteristic differentiation of each lane image, enhances impulse noise suppression on a light dense lane, enhances Gaussian noise removal on a dark area lane, analyzes brightness distribution of each lane image in real time, reduces local exposure gain on an over-bright area, improves exposure integration time on the over-dark area, ensures brightness balance of each lane image, and then transmits the images to an AI system, so that brightness interference during AI recognition is reduced.
  4. 4. The multi-lane snapshot night vision gun camera according to claim 3, wherein the multi-branch segmentation network of the AI scene segmentation recognition system comprises lane exclusive branches and global association branches, the lane exclusive branches perform semantic segmentation on each lane image independently and output local features such as vehicle position and license plates in the lanes, the global association branches fuse segmentation results of all lanes, global features such as cross-lane vehicle lane change behavior multi-lane traffic density are analyzed, meanwhile, target tracking sub-models are matched through cross frames of license plate feature vehicle appearance features, track association of the same vehicle in different time frames of different lanes is established, misjudgment of vehicle identities among multiple lanes is avoided, and the system can dynamically adjust snapshot frequency according to the global traffic density.
  5. 5. The multi-lane snapshot night vision gun camera according to claim 1, wherein the night vision full-color optimization module comprises a multi-spectrum photosensitive assembly and an AI color calibration subsystem, wherein the multi-spectrum photosensitive assembly synchronously collects visible light signals and near infrared light signals of all lanes to generate multi-spectrum image data, the AI color calibration subsystem receives lane scene characteristics output by an AI scene segmentation recognition system, if strong car light interference exists in the lanes, the fusion proportion of the near infrared light signals is automatically reduced to avoid color distortion, if the lanes are recognized as dark areas, the fusion precision of the near infrared light signals and the visible light signals is improved, the color details of the dark areas are repaired through a spectrum mapping algorithm, and meanwhile, the module adjusts color reconstruction parameters according to the difference of road surface reflectivity of all lanes.
  6. 6. The multi-lane snapshot night vision gun machine according to claim 1, wherein the snapshot cooperative control unit is internally provided with a lane priority scheduling sub-module and an image tag generation sub-module, wherein the lane priority scheduling sub-module sets a snapshot priority according to lane vehicle violation risks fed back by an AI system, the lane with high violation risks triggers the snapshot preferentially to ensure important target priority records, the image tag generation sub-module automatically adds tag information such as lane number snapshot time vehicle attributes and the like when snapshot images are stored, the tag is stored in association with image data, and the snapshot records of corresponding lanes can be rapidly screened through lane numbers when subsequent queries are performed.
  7. 7. The multi-lane snapshot night vision rifle bolt according to claim 1, further comprising an AI self-adaptive learning unit, wherein the AI self-adaptive learning unit collects snapshot image quality data and environment data of each lane in real time, trains a scene-parameter mapping model through a reinforcement learning algorithm, and when environmental changes are detected, automatically optimizes a segmentation threshold value of a preprocessing parameter AI scene segmentation recognition system of an algorithm preprocessing unit and a photosensitive parameter of a night vision full-color optimization module, so that the rifle bolt always maintains an optimal snapshot effect in different night environments.
  8. 8. A multi-lane snapshot night vision bolt face according to any of claims 1-7, comprising a mounting plate (1), a support frame (4), a main bolt face (5) and two auxiliary bolt faces (7), further comprising: The sliding groove (2) is formed in the mounting plate (1), two rectangular plates (3) are slidably mounted in the sliding groove (2), power cavities (18) are formed in the two rectangular plates (3) and the mounting plate (1), limiting blocks (19) are mounted in the three power cavities (18) in an inserted mode, the three supporting frames (4) are respectively and fixedly mounted on the three limiting blocks (19), and the main gun machine (5) and the two auxiliary gun machines (7) are respectively and fixedly mounted on the three supporting frames (4); The driving assembly is positioned in the mounting plate (1) and used for driving the two rectangular plates (3) to slide; The three groups of fixing components are respectively positioned in the mounting plate (1) and the two rectangular plates (3) and are used for fixing the positions of the four main gun units (5); The limiting groove (8) is formed in the mounting plate (1), sliding strips (9) are arranged in the limiting groove (8) in a sliding mode, a first guide column (10) and a second guide column (11) are fixedly arranged on the sliding strips (9), two first fixing strips (12) are fixedly arranged on one side of each auxiliary bolt (7), two second fixing strips (13) are fixedly arranged on one side of each main bolt (5), the first guide column (10) is located between the two second fixing strips (13), and the second guide column (11) is located between the four first fixing strips (12); The limiting assembly is positioned in the mounting plate (1) and used for limiting the position of the sliding bar (9).
  9. 9. The multi-lane snapshot night vision bolt machine of claim 8, wherein said drive assembly comprises: The motor (14), motor (14) fixed mounting just is located between two rectangular plates (3) in sliding tray (2), the output shaft fixed mounting of motor (14) has first bevel gear (15), the both sides of first bevel gear (15) are all meshed and are installed second bevel gear (16), two the one end of second bevel gear (16) is all fixed mounting threaded rod (17), two the one end of threaded rod (17) runs through two rectangular plates (3) respectively, two the screw thread on threaded rod (17) revolves to opposite directions, and two threaded rods (17) respectively with two rectangular plates (3) threaded connection, first bevel gear (15), two second bevel gears (16) and two threaded rods (17) all are connected with sliding tray (2) rotation; The fixing assembly includes: Three first fixed column (21), three equal fixed mounting of first fixed column (21) is in power chamber (18), and is three slidable mounting has same sliding block (20) on first fixed column (21), and is three all the cover is equipped with first spring (22) on first fixed column (21), two rectangular channel (23) have been seted up on sliding block (20), two equal slidable mounting has sliding column (24) in rectangular channel (23), two equal fixed mounting of one end of sliding column (24) has gag lever post (25), two in one side of gag lever post (25) all extends to stopper (19), two gag lever post (25) all peg graft the cooperation with stopper (19), sliding block (20) and power chamber (18) sliding connection, two gag lever post (25) all rotate with power chamber (18) and be connected.
  10. 10. The multi-lane snapshot night vision bolt face of claim 9, wherein said stop assembly comprises: The two second fixing columns (28), two second fixing columns (28) are fixedly arranged in the limiting grooves (8), two second fixing columns (28) are slidably provided with the same sliding plate (27), two second fixing columns (28) are sleeved with second springs (29), the bottoms of the sliding plates (27) are provided with a plurality of guide grooves (30), the guide grooves (30) are internally provided with limiting columns (31) in a sliding manner, the bottoms of the limiting columns (31) are fixedly provided with rectangular strips (32), the tops of the sliding strips (9) are fixedly provided with fixing plates (26), the fixing plates (26) are provided with a plurality of grooves (33), the rectangular strips (32) are in plug-in fit with the grooves (33), and the sliding plates (27), the three rectangular strips (32) and the fixing plates (26) are slidably connected with the limiting grooves (8); Four mounting holes (34) are formed in the mounting plate (1); the second guide posts (11) are closely contacted with four first fixing strips (12), and the first guide posts (10) are closely contacted with two second fixing strips (13); the main bolt (5) is positioned between two auxiliary bolts (7); The guide rod (6) is fixedly installed in the sliding groove (2), one end of the guide rod (6) penetrates through the two rectangular plates (3), and the two rectangular plates (3) are both in sliding connection with the guide rod (6).

Description

Night vision rifle bolt of multilane snap shot Technical Field The invention belongs to the technical field of video monitoring processing equipment, and particularly relates to a multi-lane snapshot night vision gun camera. Background The current multi-lane snapshot night vision gun camera faces a plurality of technical bottlenecks in practical application, and is difficult to meet high-efficiency and accurate snapshot requirements under complex traffic scenes at night. In the aspect of multi-lane imaging, the traditional gun camera mostly adopts a single-sensor architecture, multiple lanes are covered through a wide-angle lens, images of different lanes are easy to overlap, imaging blurring occurs to lane edge targets due to the deviation of an angle of view, and the single sensor is difficult to synchronously adapt to illumination differences of all lanes (such as that part of lanes are directly irradiated by a street lamp and part of lanes are in a shadow area), so that part of lane images are too bright or too dark, and the subsequent target recognition is influenced. From the aspect of algorithm processing, the image preprocessing and AI recognition of the existing equipment are concentrated at the rear end, the front end is only responsible for image acquisition and transmission, a large amount of raw image data which is not processed occupies transmission bandwidth, so that AI processing delay is increased, and particularly when vehicles on multiple lanes are dense at night, the delay problem is more remarkable, and snapshot missing or snapshot time deviation is easy to occur. Meanwhile, the traditional AI identification lacks of fine segmentation of a multi-lane scene, multi-lane images are processed as a whole, and independent scene characteristics (such as vehicle density and vehicle type distribution of different lanes) of each lane are difficult to distinguish, so that snapshot triggering conditions are single, and accurate adjustment cannot be performed according to actual conditions of the lanes. In the aspect of the cooperation of night vision full-color imaging and snapshot, the night vision imaging parameters of the traditional gun camera are mostly fixed and preset, cannot be dynamically adjusted according to real-time scene characteristics (such as vehicle light intensity and road surface reflectivity) of each lane, partial lane color distortion and detail loss are easy to occur, the snapshot control unit is poor in linkage with the imaging and AI recognition modules, the triggering time of a shutter is asynchronous during multi-lane snapshot, cross-lane vehicle snapshot dislocation is easy to be caused, meanwhile, the snapshot image lacks lane attribute marks, and the follow-up classification query efficiency is low. In addition, the existing equipment lacks a closed loop cooperative mechanism of imaging, preprocessing, AI identification, night vision optimization and snapshot control, and each module is functionally split, so that the equipment is difficult to cope with complex and changeable traffic environments of multiple lanes at night, and the snapshot efficiency is low, the image quality is poor and the adaptability is weak. Disclosure of Invention The invention aims to provide a multi-lane snapshot night vision gun camera, which aims to solve the problems in the background technology. The invention provides a multi-lane snapshot night-vision gun camera, which comprises a multi-lane synchronous imaging module, an algorithm pre-processing unit, an AI scene segmentation recognition system, a night-vision full-color optimization module and a snapshot cooperative control unit, wherein the multi-lane synchronous imaging module adopts a multi-sensor array architecture, each sensor corresponds to an independent lane imaging area, can synchronously acquire night image data of each lane and generate lane independent image streams to avoid target interference caused by overlapping of multi-lane images, the algorithm pre-processing unit is integrated between the imaging module and an AI system, is internally provided with a lightweight image preprocessing algorithm, carries out real-time noise reduction, edge enhancement and dynamic exposure correction on each lane independent image stream, reduces data transmission quantity and improves subsequent AI processing efficiency, the AI scene segmentation recognition system receives the pre-processed lane image data, carries out scene semantic segmentation on each lane through a multi-branch segmentation network, and simultaneously calls a target tracking sub-model to realize cross-frame vehicle track association, and combines with the AI image capture optimization module to generate accurate snapshot triggering conditions, the AI image preprocessing algorithm interacts with the image capturing system to realize the full-color image capturing system, and has no real-time window control of the full-color image capturing system, and has no real-tim