Vehicle three-dimensional detection method based on model fitting algorithm

Abstract

The invention provides a vehicle three-dimensional detection method based on a model fitting algorithm. The method mainly involves vehicle dimension estimation, vehicle model fitting and two stage refining of convolutional neural network. The process of the method is that an original image is transmitted to a two-dimensional detection network, the network generates a two-dimensional bounding box for candidate vehicles in an image plane; a set of three-dimensional points that fall into the two-dimensional bounding box after projection are selected; the model fitting algorithm is performed on detection of a three-dimensional position and the three-dimensional bounding box of a vehicle by utilizing the set; the points suitable for the three-dimensional bounding box are input; the two stage refining convolutional neural network is designed; the detected three-dimensional bounding box is aligned with point cloud; and final three-dimensional box regression and classification are performed. The model fitting algorithm proposed by the method can provide three-dimensional information by utilizing the advantages of either two-dimensional detection network, a more efficient model fitting process is achieved, and the capability and detection precision of three-dimensional vehicle detection are improved.

Description

technical field [0001] The invention relates to the field of image detection, in particular to a three-dimensional vehicle detection method based on a model fitting algorithm. Background technique [0002] With the continuous improvement of people's living standards, the number of cars is increasing year by year, and the monitoring and management of vehicles is becoming more and more difficult. Therefore, the use of intelligent identification methods to detect vehicles can greatly improve the accuracy of monitoring and management, and can also reduce a lot of manpower, material and financial resources. The detection of moving vehicles can realize the information collection and processing of moving vehicles on the road, and obtain the characteristic information such as the size and position of the vehicle and the vehicle license information. In a complete intelligent transportation system, vehicle detection can provide strong data support and information support for many tra...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems of decreased vehicle detection accuracy and difficult classification, the purpose of the present invention is to provide a three-dimensional vehicle detection method based on model fitting algorithm, the original image is sent to the two-dimensional detection network, which is the candidate vehicle in the image plane Generate a 2D bounding box, select a set of 3D points that fall into the 2D bounding box after projection, using this set, the model fitting algorithm detects the 3D position of the vehicle and the 3D bounding box, and then takes the points that fit into the 3D bounding box as input, Design a two-stage refined convolutional neural network to further align the detected 3D boxes to the point cloud for final 3D box regression and classification

Images