Behavior detection system applied to complex scene

Abstract

The invention discloses a behavior detection system applied to a complex scene. The behavior detection system is structurally characterized in that a network camera is connected with a hard disk video recorder in a field control box through a switch in the field control box; the 5G wireless signal transceiver unit CPE is connected with the switch; the detection module taking the CPU and the GPU as the core is connected with the switch; the detection module carries out sampling processing on a video stream collected by the network camera and sends a processing result to the background end through the CPE. In the detection module, an AI algorithm is adopted to process and identify a video stream, and in the AI algorithm, skeleton information of a person in each frame is extracted from a video frame sequence obtained by an image acquisition device; carrying out normalization processing on position features of skeleton joints of the skeleton information in corresponding frames; and finally, performing identification by adopting an action skeleton identification network, and performing classified output.

Description

technical field [0001] The invention relates to the technical field of target recognition, in particular to a behavior detection system applied to complex scenes. technical background [0002] The research purpose of behavior detection is to locate the specific temporal and even spatial location of behaviors of interest from long videos, which is one of the most important video understanding tasks. In complex application scenarios, the target is usually affected by factors such as illumination change, noise, environment, etc., and has a large appearance change, which leads to the degradation of algorithm performance, and it becomes a very challenging task to accurately track pedestrians. What needs to be considered is that complex scenes are changing, and pedestrian tracking needs to adapt to dynamic changes in the scene, especially changes in lighting conditions, such as changes in light intensity caused by fluttering clouds, and sudden changes in weather. [0003] With th...

AI Technical Summary

Problems solved by technology

However, using the background subtraction method to obtain the moving area in the scene, and using a layer-based framework to detect shadows, only using color features to detect shadows, can only detect large areas of shadows, and the robustness is low.

Images