Search

CN-121982632-A - Pedestrian abnormal behavior real-time early warning system based on image analysis

CN121982632ACN 121982632 ACN121982632 ACN 121982632ACN-121982632-A

Abstract

The invention relates to the technical field of abnormal behavior real-time early warning. The invention relates to a pedestrian abnormal behavior real-time early warning system based on image analysis. The system comprises an image acquisition module, an early warning starting module, a safety personnel management module, an early warning adjustment module and an early warning feedback module, wherein the image acquisition module is used for acquiring image data of a parking lot, extracting images of vehicles and pedestrians in the image data to acquire vehicle images and pedestrian images, the early warning starting module is used for identifying vehicle information according to the vehicle images and extracting vehicle door images according to the vehicle images, the scene difference of pedestrian quantity change in the parking lot is effectively adapted through the design of dynamically adjusting abnormal monitoring distance, the detection range is optimized in real time by combining the pedestrian density around the vehicle, false early warning flood when pedestrians are dense is avoided, risk omission when pedestrians are sparse is prevented, and scene suitability and accuracy of early warning judgment are greatly improved.

Inventors

  • XU SONG
  • XU WENLI
  • ZHANG SHENG

Assignees

  • 昆明铁道职业技术学院(昆明市教育对外合作交流中心)

Dates

Publication Date
20260505
Application Date
20251230

Claims (9)

  1. 1. The pedestrian abnormal behavior real-time early warning system based on image analysis is characterized by comprising an image acquisition module, an early warning starting module, a safety personnel management module, an early warning adjustment module and an early warning feedback module; The image acquisition module is used for acquiring image data of a parking lot, extracting images of vehicles and pedestrians from the image data and acquiring vehicle images and pedestrian images; the early warning starting module is used for carrying out vehicle information identification according to the vehicle image, extracting a vehicle door image according to the vehicle information in combination with the vehicle image, judging the running state of the vehicle according to the vehicle image, and starting early warning management when the vehicle is judged to stop running; The safety personnel management module is used for establishing a safety personnel database which is updated regularly according to vehicle information, extracting vehicle images of historical time nodes from image data, setting dynamic association conditions by using the vehicle door images, and adding personnel leaving the vehicle into the safety personnel database by using a change time node of the vehicle door images as a reference when the dynamic association conditions are triggered; The early warning adjustment module is used for dynamically setting an abnormal monitoring distance according to the number of the pedestrian images by taking the vehicle image as a center, and adjusting the early warning state according to the distance between the pedestrian image and the vehicle, the abnormal monitoring distance and the safety personnel database; and the early warning feedback module is used for comparing the pedestrian orientation of the pedestrian image with the vehicle door image when the monitoring adjustment is not closed, then counting the opening time according to the comparison result, setting early warning time, and simultaneously carrying out differential monitoring on the vehicle door image, wherein the time counting reaches the early warning time or the vehicle door image is different, and then feeding back an abnormal behavior early warning notice to the parking lot.
  2. 2. The pedestrian abnormal behavior real-time early warning system based on image analysis of claim 1, wherein the image acquisition module is connected with a management center of a parking lot, so that image data related to the interior of the parking lot is acquired through a camera device of the parking lot, and an image of a vehicle and a pedestrian is extracted from the image data through a target detection algorithm, so that a vehicle image and a pedestrian image in the parking lot are acquired; The method comprises the steps of locking a vehicle body, a license plate and a vehicle door area of a vehicle during vehicle image extraction, and locking the human body outline, head orientation and limb action characteristics of a pedestrian during pedestrian image extraction.
  3. 3. The pedestrian abnormal behavior real-time early warning system based on image analysis according to claim 1, wherein in the early warning starting module, vehicle information is collected through a network, then a vehicle information database is built by collecting the collected vehicle information, vehicle images are combined with the vehicle information database to identify vehicle models, and then a vehicle operation related image library and a vehicle stop operation related image library are built by screening the vehicle information database according to the vehicle models; And determining a door area of the vehicle according to the model of the vehicle.
  4. 4. The pedestrian abnormal behavior real-time early warning system based on image analysis of claim 1, wherein in the early warning starting module, the running state of the vehicle is judged according to the image of the vehicle; Acquiring a parking space position of a parking lot, setting a standing time threshold value of a vehicle, carrying out position overlapping analysis on a vehicle position corresponding to a vehicle image and the parking space position, carrying out displacement variation analysis according to the vehicle image after the vehicle position is overlapped with the parking space position, and analyzing and acquiring the displacement variation of the vehicle; When the position change quantity is 0, the vehicle image is combined with the vehicle operation related image library and the vehicle operation stop related image library to be identified, and when the identification result classifies the vehicle image into the vehicle operation stop related image library, the vehicle operation stop is judged, and the early warning management is started; When the position change amount is not 0, the vehicle image is not overlapped with the parking space position, the recognition result classifies the vehicle image into a vehicle operation related image library, one condition is met, and early warning management is not carried out.
  5. 5. The pedestrian abnormal behavior real-time early warning system based on image analysis of claim 1, wherein the safety personnel management module establishes a safety personnel database exclusive to each vehicle according to license plates of vehicle information, and the safety personnel database contains personnel facial features associated with the vehicles; The periodic updating period of the security personnel database is 24 hours, and the personnel information which is not associated with the vehicle for more than 30 days is deleted when updating.
  6. 6. The pedestrian abnormal behavior real-time early warning system based on image analysis of claim 1, wherein the safety personnel management module takes a license plate of vehicle information and 30 days as screening conditions, and combines the screening conditions to extract historical vehicle images in image data of historical event nodes, and then takes a vehicle door image as a dynamic association condition; wherein, people corresponding to the door image from the driving position are needed to appear, then the door is judged to appear to open or close according to the door image, and when the door is opened or closed, the historical vehicle image is judged to meet the triggering association condition; When no personnel appear in the vehicle door image corresponding to the driving position, judging that the historical vehicle image does not meet the triggering association condition; When the historical vehicle image triggers the association condition, setting a time node for opening or closing a vehicle door as a reference, setting a personnel acquisition period in combination with the time node for extending, acquiring facial images of personnel leaving the vehicle or entering the vehicle in the personnel acquisition period, and inputting the acquired facial images into a safety personnel database corresponding to the vehicle to realize association between the personnel and the vehicle.
  7. 7. The pedestrian abnormal behavior real-time early warning system based on image analysis of claim 1, wherein in the early warning adjustment module, a dynamic adjustment rule of an abnormal monitoring distance is set, then the number of pedestrian images around the vehicle is obtained according to the position of the vehicle for starting early warning management as the center, and the setting of the abnormal monitoring distance is completed by combining the number of pedestrian images with the dynamic adjustment rule; The number of pedestrian images around the vehicle is twenty meters, and each pedestrian image represents one pedestrian; the greater the number of pedestrian images, the smaller the anomaly monitoring distance; the smaller the number of pedestrian images, the larger the anomaly monitoring distance; Acquiring a pedestrian position according to a pedestrian image, then acquiring a human-vehicle distance by combining the pedestrian position with a vehicle position, judging the human-vehicle distance by combining the abnormal monitoring distance, identifying the pedestrian image according to a safety personnel database of a corresponding vehicle when the human-vehicle distance is smaller than the abnormal monitoring distance, and releasing early warning management of the vehicle when an identification result is displayed as a safety personnel, and recovering the early warning management after the safety personnel leaves the abnormal monitoring distance; otherwise, when the identification result shows as unsafe personnel, the early warning management of the vehicle is maintained.
  8. 8. The real-time pedestrian abnormal behavior early warning system based on image analysis of claim 1, wherein in the early warning feedback module, when the early warning adjustment module does not release early warning management, the head of the pedestrian is subjected to orientation analysis according to the pedestrian image, and when the head orientation of the pedestrian is smaller than 30 degrees with the included angle of a vehicle door and the distance between the pedestrian and the vehicle is gradually reduced, the time is started; When the distance between the person and the vehicle is smaller than 1m, directly starting time counting; when the head orientation of the pedestrian is greater than 30 degrees with the vehicle door, the time counting is not started.
  9. 9. The pedestrian abnormal behavior real-time early warning system based on image analysis of claim 8, wherein after the timing is started, the early warning time is set, and meanwhile, difference monitoring is carried out on the vehicle door image; When the timing time exceeds the early warning time, judging that the pedestrian behavior is abnormal, and feeding back an abnormal behavior early warning notice to a management center of the parking lot in combination with the pedestrian position; And carrying out differential analysis on the door images of different time nodes to obtain differential data of the door images, wherein the differential data comprises the width variation of the door gap and the differential state of the door lock, and when the width variation of the gap exceeds 5mm or the door lock state is changed from closed to open, judging that the behavior of a pedestrian is abnormal, and feeding back an abnormal behavior early warning notice to a management center of a parking lot in combination with the position of the pedestrian.

Description

Pedestrian abnormal behavior real-time early warning system based on image analysis Technical Field The invention relates to the technical field of abnormal behavior real-time early warning, in particular to a pedestrian abnormal behavior real-time early warning system based on image analysis. Background In parking lot management, vehicle safety protection and pedestrian behavior monitoring are core requirements for guaranteeing the operation order of a parking lot and reducing the property loss of vehicles. The existing pedestrian abnormal behavior monitoring technology based on image analysis mostly adopts fixed abnormal monitoring distance, does not consider scene difference of dynamic change of the number of pedestrians in a parking lot, when pedestrians are dense, a fixed long-distance threshold value is easy to cause false warning and floods, when pedestrians are sparse, a fixed short-distance threshold value is likely to miss risk behaviors close to the long distance, accuracy and practicality of warning are seriously affected, automobiles are lack of accurate distinction of pedestrian identities in the prior art, pedestrians close to vehicles are uniformly brought into a monitoring range, no special exemption mechanism is established for owners and related safety personnel, a large number of behaviors such as getting on and getting off normally are misjudged as abnormal, invalid workload of management personnel is increased, response efficiency to real risk behaviors is reduced, and in order to reduce the situation, a pedestrian abnormal behavior real-time warning system based on image analysis is provided. Disclosure of Invention The invention aims to provide a pedestrian abnormal behavior real-time early warning system based on image analysis so as to solve the problems in the background technology. In order to achieve the above purpose, the pedestrian abnormal behavior real-time early warning system based on image analysis is provided, and comprises an image acquisition module, an early warning starting module, a security personnel management module, an early warning adjustment module and an early warning feedback module; The image acquisition module is used for acquiring image data of a parking lot, extracting images of vehicles and pedestrians from the image data and acquiring vehicle images and pedestrian images; the early warning starting module is used for carrying out vehicle information identification according to the vehicle image, extracting a vehicle door image according to the vehicle information in combination with the vehicle image, judging the running state of the vehicle according to the vehicle image, and starting early warning management when the vehicle is judged to stop running; The safety personnel management module is used for establishing a safety personnel database which is updated regularly according to vehicle information, extracting vehicle images of historical time nodes from image data, setting dynamic association conditions by using the vehicle door images, and adding personnel leaving the vehicle into the safety personnel database by using a change time node of the vehicle door images as a reference when the dynamic association conditions are triggered; The early warning adjustment module is used for dynamically setting an abnormal monitoring distance according to the number of the pedestrian images by taking the vehicle image as a center, and adjusting the early warning state according to the distance between the pedestrian image and the vehicle, the abnormal monitoring distance and the safety personnel database; and the early warning feedback module is used for comparing the pedestrian orientation of the pedestrian image with the vehicle door image when the monitoring adjustment is not closed, then counting the opening time according to the comparison result, setting early warning time, and simultaneously carrying out differential monitoring on the vehicle door image, wherein the time counting reaches the early warning time or the vehicle door image is different, and then feeding back an abnormal behavior early warning notice to the parking lot. As a further improvement of the technical scheme, in the image acquisition module, connection is established with a management center of the parking lot, so that image data related to the interior of the parking lot is acquired through a camera device of the parking lot, and an image of a vehicle and a pedestrian is extracted from the image data through a target detection algorithm, so that a vehicle image and a pedestrian image in the parking lot are acquired; The method comprises the steps of locking a vehicle body, a license plate and a vehicle door area of a vehicle during vehicle image extraction, and locking the human body outline, head orientation and limb action characteristics of a pedestrian during pedestrian image extraction. As a further improvement of the technical scheme, in the early warning starting module,