作者: D Gallup , S. J. Kim , Jan-Michael Frahm , A Akbarzadeh , Q Yang
DOI:
关键词: Artificial intelligence 、 Computer science 、 Dynamic range 、 Graphics processing unit 、 Stereo camera 、 Central processing unit 、 Key (cryptography) 、 Brightness 、 Feature (computer vision) 、 Computer vision 、 Pipeline (computing)
摘要: This paper introduces a multi-view stereo matcher that gen- erates depth in real-time from monocular video stream of static scene. A key feature our processing pipeline is it estimates global camera gain changes the tracking stage and eciently compensates for these without impacting performance. very important outdoor applications where brightness range often far exceeds dynamic camera. Real-time per- formance achieved by leveraging power graphics unit (GPU) addition to CPU. We demonstrate eectiveness approach on videos urban scenes recorded vehicle-mounted with auto-gain enabled.