Street front order event video detection method based on deep learning and motion consistency

Abstract

The invention discloses a street front order event video detection method based on deep learning and motion consistency, and relates to the artificial intelligence field and the machine vision field.The street front order event video detection method based on deep learning and motion consistency includes the steps: 1) an algorithm framework; 2) target detection; 3) calculation of motion consistency; and 4) event determination. Compared with the prior art, the street front order event video detection method based on deep learning and motion consistency can unitedly determine the events throughmultiple conditions, can design a detection of out-store business events and roadside stall business events and can accurately and quickly complete automatic detection of events, by designing a target detection deep learning network, training a scene identification model, calculating the motion information in the scene and analyzing the behavior state of the target, and by means of the mode integrating the target detection technology in static video frames and the target behavior analysis technology in dynamic videos in the video intelligent analysis field.

Description

technical field [0001] The invention relates to the fields of artificial intelligence and computer vision, and is an intelligent detection method for street order events based on image processing technology and video analysis technology and applied in urban monitoring scenes. Background technique [0002] Wen Jun, in his doctoral thesis "Research on Outdoor Scene Understanding Based on Deep Convolutional Neural Network" in March 2016, disclosed that based on the DCNN algorithm, he studied dynamic object classification, semantic segmentation and joint object detection around scene segmentation and scene recognition. Scene Understanding Techniques with Semantic Segmentation. [0003] Firstly, for the classification of moving target objects in videos, a dynamic target classification method based on multi-task spatial pyramid pooling DCNN is proposed. High-level convolutional features are robust to translation, viewing angle changes, lighting, partial occlusion, etc. of moving ...

AI Technical Summary

Problems solved by technology

[0011] The problem with the above-mentioned traditional algorithms is that in terms of identifying and understanding urban monitoring scenes, traditional algorithms are difficult to meet the needs.

These factors require the algorithm to have super generalization ability and accuracy. Traditional algorithms cannot meet these requirements on a theoretical basis. Even if they are applied, it is easy to cause false detection and missed detection of events.

[0012] The problem with the above algorithm based on deep learning is that based on the deep learning algorithm, a method with strong generalization ability can be designed to solve the problem of various event forms, but it also has high requirements for the design of the network model

Therefore, the simple application of static information of image frames has low accuracy and many false detections and missed detections in target detection and scene recognition.

Images