Multiscale object detection device and method
View Patent ↗There is provided a multi-scale object detection device. The device includes an image frame acquisition unit for acquiring a plurality of consecutive image frames, a critical region extractor for extracting at least one second critical region from a current image frame based on at least one first critical region extracted from a previous image frame among the consecutive image frames, a multi-scale object detector whose operation involves a first object detection process for the current image frame and a second object detection process for the at least one second critical region, and an object detection integration unit for integrating the results of the first and second object detection processes.
1 . A multi-scale object detection device comprising:
an image frame acquisition unit for acquiring a plurality of consecutive image frames;
a critical region extractor for extracting at least one second critical region from a current image frame based on at least one first critical region extracted from a previous image frame among the consecutive image frames;
a multi-scale object detector whose operation involves a first object detection process for the current image frame and a second object detection process for the at least one second critical region; and
an object detection integration unit for integrating results of the first and second object detection processes,
wherein the critical region extractor
determines a position of a reference object of the at least one first critical region on the previous image frame, matches the reference object to an object of the current image frame, and determines a position of the at least one second critical region as a position of the object of the current image frame, and
defines the second critical region as a region at the position of the object of the current image frame and having a boundary set at a distance from the position of the object of the current image frame that corresponds to a distance from the reference object of the at least one first critical region to a boundary of the at least one first critical region, and
wherein the image frame acquisition unit, the critical region extractor, the multi-scale object detector, and the object detection integration unit are each implemented via at least one processor.
2 . The multi-scale object detection device of claim 1 ,
wherein the image frame acquisition unit analyzes sequential video frames and extracts image frames that are successive in time series and have a similar background to determine the plurality of consecutive image frames.
3 . The multi-scale object detection device of claim 1 ,
wherein the at least one first critical region is fed back to the critical region extractor by the multi-scale object detector.
4 . The multi-scale object detection device of claim 1 ,
wherein the multi-scale object detector detects a first object after down-sampling the current image frame in the first object detection process.
5 . The multi-scale object detection device of claim 4 ,
wherein the multi-scale object detector determines a position and a class of the first object by detecting the first object.
6 . The multi-scale object detection device of claim 4 ,
wherein the multi-scale object detector extracts a position and a class of a second object with a relatively high accuracy from the at least one second critical region in the second object detection process.
7 . The multi-scale object detection device of claim 1 ,
wherein the multi-scale object detector assigns the first object detection process to a first processor and assigns the second object detection process to a second processor to process the first and second object detection processes in parallel.
8 . The multi-scale object detection device of claim 1 ,
wherein the object detection integration unit detects objects on the entire current image frame by integrating a second object extracted in the second object detection process with a first object extracted in the first object detection process.
9 . A multi-scale object detection method comprising:
an image frame acquisition step of acquiring a plurality of successive image frames;
a critical region extraction step of extracting at least one second critical region from a current image frame based on at least one first critical region extracted from a previous image frame among the successive image frames;
a multi-scale object detection step involving a first object detection process for the current image frame and a second object detection process for the at least one second critical region; and
an object detection integration step of integrating results of the first and second object detection processes,
wherein the critical region extraction step involves a step in which
a position of a reference object of the at least one first critical region on the previous image frame is determined, the reference object is matched to an object of the current image frame, and a position of the at least one second critical region is determined as a position of the object of the current image frame, and
the second critical region is defined as a region at the position of the object of the current image frame and having a boundary set at a distance from the position of the object of the current image frame that corresponds to a distance from the reference object of the at least one first critical region to a boundary of the at least one first critical region.
10 . The multi-scale object detection method of claim 9 ,
wherein the critical region extraction step involves a step in which the at least one first critical region determined in the multi-scale object detection step is fed back.
11 . The multi-scale object detection method of claim 9 ,
wherein the multi-scale object detection step involves a step in which a first object is detected after down-sampling the current image frame in the first object detection process.
12 . The multi-scale object detection method of claim 11 ,
wherein the multi-scale object detection step involves a step in which a position and a class of a second object are extracted with a relatively high accuracy from the at least one second critical region in the second object detection process.
13 . The multi-scale object detection method of claim 9 ,
wherein the multi-scale object detection step involves a step in which the first object detection process is assigned to a first processor and the second object detection process is assigned to a second processor so that the first and second object detection processes are processed in parallel.