PROBLEM TO BE SOLVED: To provide a vehicular pedestrian image acquisition system capable of acquiring an image of a pedestrian, who exists in the vicinity of a vehicle, as a raw material for a learning sample, which is used to perform machine learning and is as precise as an original image, while taking account of protection of a subject's portrait right or privacy.SOLUTION: A vehicular pedestrian image acquisition system (1) includes a camera (2) that images the vicinity of a vehicle, a face area detection unit (6) that detects the position and size of a face area of a pedestrian contained in a produced vehicle vicinity image, a feature quantity calculation unit (8) that calculates a feature quantity for each face area on the basis of information on the face area of each pedestrian in the vehicle vicinity image, a non-reversible transformation processing unit (10) that performs non-reversible transformation on the face area of each pedestrian in the vehicle vicinity image, and a pedestrian image preservation unit (12) that associates information, which specifies the position and size of the face area of each pedestrian in the vehicle vicinity image, with a feature quantity calculated for the face area, and stores the information and feature quantity together with the vehicle vicinity image in which the face area of each pedestrian is subjected to the non-reversible transformation.SELECTED DRAWING: Figure 1
【課題】被写体の肖像権やプライバシーの保護に配慮しつつ元の画像と同等の精度を確保した機械学習を行うための学習用サンプルの素材として、車両周辺に存在する歩行者画像を取得することができる、車両用歩行者画像取得装置を提供する。【解決手段】車両用歩行者画像取得装置(1)は、車両の周辺を撮影するカメラ(2)と、撮影された車両周辺画像中に含まれる歩行者の顔領域の位置及び大きさを検出する顔領域検出部(6)と、車両周辺画像における各歩行者の顔領域内の情報に基づき、その顔領域毎に特徴量を算出する特徴量算出部(8)と、車両周辺画像における各歩行者の顔領域を非可逆変換する非可逆変換処理部(10)と、車両周辺画像における各歩行者の顔領域の位置及び大きさを特定する情報とそれらの各顔領域について算出された特徴量とを相互に関連付けて、各歩行者の顔領域を非可逆変換した車両周辺画像と共に保存する歩行者画像保存部(12)とを有する。【選択図】図1
VEHICULAR PEDESTRIAN IMAGE ACQUISITION SYSTEM
車両用歩行者画像取得装置
2016-07-11
Patent
Electronic Resource
Japanese
VEHICULAR PEDESTRIAN COLLISION SENSING UNIT AND VEHICULAR PEDESTRIAN COLLISION SENSING PROGRAM
European Patent Office | 2015
|