[CHEN, Yanyi 陈延艺, XIA, Qiming 夏启明, DU, Jing 杜静, HUANG, Shangfeng 黄尚锋, CHEN, Yanxing 陈延行, JIANG, Wentao 江文涛] Unit 102, 59 Wanghai Road, Software Park Phase II, Siming District 中国福建省厦门市思明区软件园二期望海路59号102单元, Fujian 361000Xiamen, Fujian 361000;102-3, 59 Wanghai Road, Software Park Phase II, Siming District 中国福建省厦门市厦门市思明区软件园二期望海路59号102之三, Fujian 361000Xiamen, Fujian 361000 Disclosed are a three-dimensional heat map-based three-dimensional point cloud target detection method and device. Said method comprises: inputting a point cloud into a multi-layer sparse convolution to obtain first feature maps, using a bilinear interpolation method to sequentially interpolate the first feature maps to an original spatial position of the point cloud to obtain second feature maps, and splicing the second feature maps to output an N * C-dimensional feature map; inputting the N * C-dimensional feature map into a first fully-connected neural network, to output an N * 4-dimensional feature map, wherein the output data comprises the predicted coordinates of the point cloud and a predicted thermal response value; inputting the N * C-dimensional feature map into a second fully-connected neural network, to output a five-dimensional vector, which respectively represents the predicted length, width, height, category and deflection angle of the object; performing regression on the predicted thermal response value by means of the three-dimensional Gaussian distribution to obtain a mean value of the Gaussian distribution, i.e. the predicted center coordinates of the object; and combining the length, width, height, category, deflection angle and the center coordinates to obtain a target detection result. The present disclosure can improve the detection precision and classification capability in respect of a target.

[ZHANG, Xiang 张翔, HUANG, Shangfeng 黄尚锋, DU, Jiing 杜静, XIA, Qiming 夏启明, CHEN, Yanxing 陈延行, JIANG, Wentao 江文涛] Unit 102, 59 Wanghai Road, Software Park Phase II, Siming District 中国福建省厦门市思明区软件园二期望海路59号102单元, Fujian 361000xiamen, Fujian 361000;102-3, 59 Wanghai Road, Software Park Phase II, Siming District 中国福建省厦门市思明区软件园二期望海路59号102之三, Fujian 361000xiamen, Fujian 361000 The present disclosure relates to the technical field of computers, and specifically to a method and system for assisting point cloud-based object detection. The method and system for assisting point cloud-based object detection provided in embodiments of the present application have the following advantages: 1. the number of sampled points processed is reduced, and the calculation speed is increased; 2. according to the feature of an intermediate point of an initial target point cloud set, a fully connected layer is used to output the probability that each sampled point in the initial target point cloud set is a foreground point or a background point, so that the final semantic segmentation effect is more obvious, and then the display result of point cloud-based object detection is more obvious; 3. a precise three-dimensional object bounding box position can be obtained, and thus the method achieves high precision.