A 3D Object Detection Method Based on Data Fusion

Abstract

The present invention provides a 3D target detection method based on data fusion, which is realized by using a deep convolutional network, and specifically includes: firstly, a feature extraction method is proposed, which extracts features from a point cloud bird's-eye view and an image of a target to be detected; Then use the Markov random field model (MRF) to encode the 3D point cloud data into a global energy function using spatial constraints to extract 3D candidate proposal boxes; finally, a data fusion method is proposed to fuse multi-modal data. Classification and regression of target boxes are performed. The 3D target detection method based on data fusion proposed by the present invention can effectively improve the detection and positioning performance of the detection network for different targets of interest in 3D space in different environments, and solve the problem of pedestrians caused by point cloud sparsity in the road environment. and poor vehicle detection.

Description

technical field [0001] The invention belongs to the fields of image processing and computer vision, and in particular relates to a 3D target detection method based on data fusion. Background technique [0002] With the vigorous development of artificial intelligence technology, smart cars with Advanced Driving Assistant System (Advanced Driving Assistant System, ADAS) and driverless technology as the core become the development direction of future cars, and three-dimensional (3D) goals as one of its key technologies Detection has always been a research hotspot in this field. [0003] For 3D object detection, there are currently three mainstream methods: the first method is 3D object detection based on monocular RGB images, such as by X.Chen et al. (Chen X, Kundu K, Zhang Z, et al.Monocular 3D ObjectDetection for Autonomous Driving[C] / / 2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).IEEE, 2016) proposed a monocular image target detection method, which fo...

AI Technical Summary

Problems solved by technology

[0003] For 3D object detection, there are currently three mainstream methods: the first method is 3D object detection based on monocular RGB images, such as by X.Chen et al. (Chen X, Kundu K, Zhang Z, et al.Monocular 3D ObjectDetection for Autonomous Driving[C] / / 2016IEEE Conference on Computer Vision and Pattern Recognition(CVPR).IEEE, 2016) proposed a monocular image target detection method, which focuses on target shape prior, contextual features and instance segmentation of monocular images , in order to generate a 3D object suggestion box, because it is a monocular image, this method inevitably lacks accurate depth information

Images