Vehicle feature deep learning identification trajectory tracking method based on image system

Abstract

The invention provides a vehicle feature deep learning identification trajectory tracking method based on an image system. The system comprises a road monitoring camera, a calculation processing host and a display screen. The method comprises the following steps: firstly, performing deep learning algorithm target recognition on an image acquired by a camera to obtain an original image vehicle target frame data set, and then performing space projection transformation on the original image to obtain a projection image vehicle target frame data set; and associating the projection image vehicle target by Kalman filtering and a Hungary algorithm so as to generate a projection image vehicle target track. The invention provides a vehicle tracking method which can be installed on the ground and is beneficial to modeling of target motion by Kalman filtering.

Description

technical field [0001] The invention belongs to the technical field of vehicle detection, and in particular relates to a vehicle feature deep learning recognition track tracking method based on an image system. Background technique [0002] With the rapid development of vehicle recognition technology, especially the vehicle recognition technology based on image deep learning has become the mainstream, the field of multi-target tracking has also expanded new design processes. Generally speaking, the multi-target tracking technology based on image deep learning is roughly divided into two steps: first, use the target recognition algorithm to identify and extract target information, and then use Kalman filtering and data association technology to associate and match the target bounding boxes under continuous video frames. So as to achieve the purpose of target tracking target. However, the video data shot in this kind of process application scene is often shot from a squint pe...

AI Technical Summary

Problems solved by technology

However, the video data shot in this kind of process application scene is often shot from a squint perspective. In this case, the target motion state in the video is affected by the distance of the shooting distance, which increases the uncertainty of the target motion speed. At the same time, the Kalman Filtering needs to model the target's motion state. For the target's motion state under the squint perspective, it will be difficult to evaluate its motion state and then model it. Invention patent CN110675431B and invention patent applications CN112098993A and CN112070807A did not pay attention to the application of Kalman filter Scene target modeling state, invention patents CN110673620B, CN110706266B and invention patent application CN112132862A use drones to track target trajectories from a high-altitude perspective, but the flying conditions of drones are limited, and it is not conducive to popularization

Images