Visual perception method and device of curling ball, computer equipment and storage medium
A visual perception and curling technology, applied in computing, image data processing, instruments, etc., can solve problems such as misalignment of actual centers of curling balls, lack of curling ball depth and attitude information, etc., to improve accuracy and strong applicability , The effect of high-precision pose estimation
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
specific Embodiment 1
[0060] according to Figure 1-Figure 2 As shown, the present invention provides a kind of visual perception method of curling ball, comprises the following steps:
[0061] Step 1: Based on the pose estimation training data generated in the simulation environment, build and train a pose estimation network to predict the pose of the curling ball in the camera coordinate system;
[0062] Step 2: Build the curling ball pose estimation network structure, and reconstruct the segmented image corresponding to the input curling ball color image. The segmented image only contains two categories of curling ball and background, regardless of any input curling ball image. Occlusion, noise, background and lighting conditions, this structure enables the network to learn the shape of the curling ball and automatically filter out the interference information. On this basis, the added regression layer takes the feature vector in the middle of the encoder-decoder as input, Regression of 3D posi...
specific Embodiment 2
[0065] The difference between the second embodiment of the present application and the first embodiment is only:
[0066] The step 1 is specifically:
[0067] Step 1.1: Set up a curling simulation environment, and place curling balls in five rows and five columns in equal intervals at 0.5m intervals in the base camp area. Combined with the domain randomization mechanism, the curling position, camera pose, light direction and Strength, obstacles, and target materials are randomly adjusted to obtain the automatically labeled curling ball pose estimation simulation training data;
[0068] Step 1.2: Obtain the bounding box information of the curling ball in the color image; set the curling ball information as [x b the y b w b h b ] T , the original bounding box information is [x 1 the y 1 x 2 the y 2 ] T , where (x 1 ,y 1 ) and (x 2 ,y 2 ) are the image coordinates of the upper left and lower right of the bounding box respectively, W and H are the width and hei...
specific Embodiment 3
[0084] The difference between the third embodiment of the present application and the second embodiment is only:
[0085] The step 2 is specifically:
[0086] Step 2.1: Establish the encoder module, the input is a color image with a shape of (3,128,128), and the convolution layer with a step size of 2 and a kernel size of 5 is used to downsample the image for feature extraction of the input, and the compression is Dimensions are 128 eigenvectors;
[0087] Step 2.2: Establish a bounding box feature embedding module, the input is the bounding box information of the corresponding target, including the normalized [x b the y b w b h b ] T , respectively corresponding to the horizontal and vertical coordinates of the center of the target bounding box and the width and height of the bounding box;
[0088] After the fully connected layer outputs a feature vector with a dimension of 128, the feature vector is spliced with the 128-dimensional feature vector output by the enco...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More