Object detection method and device

Abstract

The invention discloses an object detection method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the steps of performing voxel processing on a space occupied by point cloud data to divide the space into at least one voxel; for each voxel, obtaining the similarity between the voxel and the target object according to the point cloud information of the voxel and the point cloud information of the preset target object; and taking the target object whose similarity with the voxel satisfies a preset threshold as an object corresponding tothe voxel so as to realize object detection. According to the embodiment of the invention, while the detection precision is improved, the data processing time and resources required by calculation aregreatly reduced, and the safety and reliability of unmanned driving are improved.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to an object detection method and device. Background technique [0002] With the rapid development and progress of artificial intelligence technology in recent years, assisted driving and automatic driving have been further realized, and their application scenarios have continued to expand. In the field of autonomous driving and unmanned driving, laser radar (Lidar) is widely used as the main sensor to detect spatial environment information. In the early days, lidar was also used in scientific research and engineering work such as tunnel engineering detection and cultural relic scanning and restoration due to its high precision and insusceptibility to external interference. Compared with cameras or other types of cameras, lidar is less susceptible to light and weather interference, and the accuracy of lidar with coordinate correction and registration can reach millimeter ...

AI Technical Summary

Problems solved by technology

[0005] 1. There is too little input information. Since most of the existing technologies are in the testing ground stage, the simulated road traffic conditions are relatively ideal and simple. Only relying on image recognition cannot cope with the detection of other objects in real complex scenes;

[0006] 2. The processing process is cumbersome and complicated. In order to accurately identify the target object, not only the captured pictures need to be input for image recognition, but also point cloud data and other information need to be input to determine the position information of the object. Therefore, a lot of data needs to be input and a large amount of calculation is required. ;

[0007] 3. The error of the calculation result is large. The calculation process of image recognition, position coordinate confirmation and other technologies takes a long time and requires more processing. It is difficult to adapt to the short delay or even real-time requirements of automatic driving or assisted driving

[0008] In summary, the existing technical solutions are mostly limited by equipment or storage and computing capabilities, and it is difficult to deal with complex environments and actual road conditions, resulting in hidden dangers in the driverless safety chain.

Images