Road scene other vehicle detection system and method based on space-time domain multi-dimensional fusion

Abstract

The invention provides a road scene other vehicle detection system and method based on a space-time domain multi-dimensional fusion technology, and the method comprises the steps: obtaining a projection matrix through employing calibration parameters of a camera and a radar, fusing radar point cloud data and camera image data through the projection matrix according to the distances and different tones, and forming multi-dimensional fusion image data; respectively detecting the camera image data and the multi-dimensional fusion image data by adopting two YOLOv3 models, calibrating a detection frame of the target vehicle, forming a first annotation picture and a second annotation picture, and performing training to obtain a first target detection model and a second target detection model; carrying out tracking and data association on the detection frame of the target vehicle, and outputting a detection result after screening. Target tracking is achieved through data association in combination with the time domain information of the image sequence frames to perform other vehicle detection, so that the method is closer to an actual road scene, the influence on the other vehicle detection precision caused by the mobility of vehicles in the road scene is reduced, and the detection precision and robustness are improved.

Description

technical field [0001] The present invention relates to the technical field of pattern recognition and artificial intelligence, in particular to a system and method for detecting other vehicles in road scenes based on space-time domain multi-dimensional fusion, especially a method that combines camera image information, radar point cloud information and target tracking A road scene other vehicle detection method based on space-time domain multi-dimensional fusion technology of information. Background technique [0002] With the continuous development of computer technology, deep learning technology has made great progress in the field of target detection. At the same time, the number of cars is constantly increasing, and people have growing new requirements for the convenience and safety of transportation tools. . In target detection, vehicle target detection is an important field. Whether the vehicle can be detected quickly and safely is the premise of the feasibility and ...

AI Technical Summary

Problems solved by technology

Such a detection method has several disadvantages. First, in the actual road scene, the self-driving vehicle has multiple sensors, such as lidar, color camera, GPS system, etc., and a single image target detection method cannot make full use of these roads. Multi-dimensional information on scene airspace

Second, the image in the road scene is in the form of a frame-by-frame continuous video sequence. In these sequences, there are a lot of redundant information and related information between image frames. A single image target detection method cannot fully Utilize the multidimensional information in the temporal domain of these road scenes

Third, on a moving vehicle, the images collected by the camera often have problems such as occlusion, out-of-focus, motion blur, and illumination changes. If a single image target detection method is adopted, these problems cannot be fully avoided and affect the accuracy of the target detection algorithm. , which brings greater difficulty to the target detection of road scenes

Accurate detection and classification of targets based on deep learning, but not suitable for multi-dimensional road scene detection

Using monitoring equipment in the traffic system, compared with urban traffic equipped with monitoring equipment, it is not applicable to road scenarios such as rural streets without monitoring equipment or lack of monitoring equipment

Images