DVR-Scan v1.5

Release Notes

• Significant performance improvements on multicore systems
• Support wildcards/globs as inputs for scanning entire folders (-i folder/*.mp4)
• Allow use of ffmpeg for better output quality (-m ffmpeg) or codec-copying mode (-m copy)
• Configuration files are now supported, see documentation for details
  • Can specify config file path with -c/--config, or create a dvr-scan.cfg file in your user config folder
• Windows binaries are now signed, thanks SignPath.io (certificate by SignPath Foundation)
• Experimental Nvidia CUDA® support has been added (set -b MOG2_CUDA)
  • If installing via pip, requires manual installation of OpenCV compiled with CUDA support
  • If downloading Windows version, make sure to download the GPU-enabled build (dvr-scan-1.5-win64-cuda.zip)
  • CUDA-enabled builds only support GTX 900-series and above GPUs, do not include the CNT method, and are not yet code signed
• Minimum supported Python version is now 3.7
• Minimum supported OpenCV version is now 3.x

Command-Line Changes

• New options:
  • -c/--config - set path to config file
  • -d/--output-dir - set directory to write output files (default is working directory)
  • -m/--output-mode - set output mode (one of: opencv, ffmpeg, copy)
  • -mo/--mask-output - path to write motion mask for analysis
  • --verbosity, --logfile - control output verbosity, path to save log output
• -i/--input now supports globs/wildcards to scan entire folders, e.g. -i folder/*.mp4
• Change default value for -l/--min-event-length to 0.1 seconds, previously was 2 frames
• Long form of -roi has been renamed to --region-of-interest (previously was --rectangle-of-interest)
• -c is now used for --config, previously was for --codec
• Add experimental MOG2_CUDA option for -b/--bg-subtractor
• Rename existing MOG option to MOG2
• --codec has been removed, the value should now be set using config file option opencv-fourcc

Changelog

• [feature] Configuration file support and new -c/--config argument to specify path to config files (#77)
  • Breaks existing behavior of -c (was previously the shortform of --codec)
• [feature] Add support for multiple output modes via -m/--output-mode argument (#27, #42)
• [feature] Experimental support for GPU-based CUDA MOG2 filter (#12)
• [feature] Video encoding and decoding are now done in parallel with the scanning logic leading to improved performance on most systems (#52)
• [feature] Add support for exporting motion masks via -mo/--mask-output argument
  • Useful for detailed analysis or tuning of detection parameters
  • ffmpeg can be used to generate output videos by specifying -m ffmpeg
  • Codec-copy mode, using ffmpeg, can be used by specifying -m copy
• [feature] Add --verbosity and --logfile arguments to provide more control over program output
• [feature] Allow scanning entire folders using wildcards with -i/--input (#5)
  • Glob expansion is also performed on each input path directly, so quoted globs also function correctly
• [bugfix] Fix incorrect results when -st/--start-time is set
• [bugfix] Event start times are now correctly calculated when using -fs/--frame-skip (#68, #70)
  • Note that all skipped frames within the event window are included in motion event, thus the calculated start time may be slightly earlier
• [bugfix] Only get screen resolution when required (was causing issues on headless machines)
• [bugfix] Fix output messages conflicting with progress bar shown during scanning
• [bugfix] Output events now start from 1 to align with the event list
• [bugfix] Event end times now include the presentation duration of the last frame
• [bugfix] Small values for -l/--min-event-length are now handled correctly, previously would cause an error
• [enhancement] Progress bar now indicates how many events have been detected so far
• [enhancement] Change default value for min_event_len to 0.1 seconds, previously was 2 frames (#91)

Known Issues

• Attempting to open multiple input videos will cause DVR-Scan to crash, fixed in v1.5.1 (#95)
• Variable framerate videos (VFR) are not fully supported, and will yield incorrect timestamps (#20)
• Video output when using frame skip and -m opencv (default output mode) will result in frames missing from the exported videos (#81)
  • Use -m ffmpeg or -m copy as a workaround
• Multiple input videos are not supported with -m ffmpeg or -m copy (#86)
  • Use ffmpeg to concatenate/merge input videos before processing as a workaround
• CUDA builds do not include the CNT option for -b/--bg-subtractor