Method for detecting persons using 1D depths and 2D texture

摘要

A method detects an object in a scene by first determining an active set of window positions from depth data. Specifically, the object can be a person. The depth data are acquired by a depth sensor. For each, window position perform the following steps. Assign a window size based on the depth data. Select a current window from the active set of window positions. Extract a joint feature from the depth data and texture data for the current window, wherein the texture data are acquired by a camera. Classify the joint feature to detect the object. The classifier is trained with joint training features extracted from training data including training depth data and training texture data acquired by the sensor and camera respectively. Finally, the active set of window positions is updated before processing the next current window.

主权项

1. A method for detecting an object including a human in a scene in real time, comprising steps:
training a classifier with joint training features extracted from training data including training one dimensional (1D) depth data acquired by a LIDAR (Light Detection And Ranging) sensor from range scans and training two dimensional (2D) texture data acquired by a camera configured to provide a three dimensional (3D) structure of the scene, so some range scans are obtained from a depth image by converting the 1D depth data into LIDAR-like readings, synthetically, wherein training the classifier includes integrating photometric and range scan features obtained from the 1D depth data and the 2D texture data, along with combining the 1D depth data and the 2D texture data via a geometric descriptor into a single joint feature, so as to construct the classifier that accepts unrestricted range scan positions on the object including a human body; determining an active set of window positions from depth data, wherein the depth data are acquired by the LIDAR sensor, such that the LIDAR sensor provides a single, horizontal, synchronously acquired range scan segment within each window and the camera provides multiple horizontal range scan segments within each window, and for each window position in the active set of window positions further comprising: assigning a window size from the depth data; selecting a current window from the active set of window positions; extracting a joint feature from the depth data and texture data for the current window, wherein the joint feature includes a depth feature concatenated with a visual feature, wherein the visual feature includes histograms of gradients (HOG); classifying the joint feature, using the classifier, to detect the object; and updating the active set of window positions before processing a next current window, wherein the steps are performed in a processor.