Target detection method, device and equipment for automatic driving

Abstract

The present application relates to a target detection method, device and equipment for automatic driving. The method includes: acquiring an image collected by a monocular shooting device; inputting the image into a preset monocular 3D target detection model to obtain the predicted height of the target in the image; according to the predicted height, the target The position of the central point of the object in the image, the installation height of the monocular shooting device in the world coordinates and the internal reference of the monocular shooting device, calculate the depth prior value of the central point of the target; according to the The depth prior value is used to obtain the depth value of the center point of the target object. The solution provided by this application can decouple the model from the monocular camera during the process of predicting the depth value z, thereby improving the generalization of the model and improving the reliability of the target detection result.

Description

technical field [0001] The present application relates to the technical field of navigation, and in particular to an object detection method, device and equipment for automatic driving. Background technique [0002] Target detection is a traditional task in the field of computer vision. Different from image recognition, target detection not only needs to identify the objects existing on the image, give the corresponding category, but also needs to pass the position of the object through the minimum bounding box (Bounding box) way give. According to the different output results required for target detection, the method of using RGB images for target detection, outputting object categories and the smallest bounding box on the image is called 2D target detection, and RGB images, RGB-D depth images and lasers will be used Point cloud, the detection of the output object category and the length, width, height, rotation angle and other information in three-dimensional space is cal...

AI Technical Summary

Problems solved by technology

[0004] However, the current monocular 3D object detection model, for the prediction of the depth value z, uses the mean and variance statistically obtained from the label information of the training set

The above mean and variance can be understood as the weight parameters in the monocular 3D target detection model applicable to the corresponding monocular camera, when the corresponding monocular camera is replaced, or the image of the input model is collected by other different monocular cameras , will affect the predicted depth value z output by the model, resulting in a decline in the correctness of the prediction of the depth value z

In other words, the current monocular 3D target detection model has poor model generalization, which in turn affects the reliability of the final detection results.

Images