Object detection device and object detection method
An image feature map generating unit ( 3 ) generates, on the basis of feature amounts extracted from a plurality of images successively captured by a camera ( 109 ), an image feature map which is an estimated distribution of the object likelihood on each of the images. An object detecting unit ( 4 ) detects an object on the basis of the image feature map generated by the image feature map generating unit ( 3 ).
1. An object detection device, comprising:
a processor to execute a program; and
a memory to store the program which, when executed by the processor, performs processes of,
acquiring, from a plurality of frame images arranged in a time series in an image sequence successively captured by a camera, a pair of same frame images for each of the plurality of frame images and inverting one of the same frame images, generating image pyramids each of which includes images obtained by gradually reducing a corresponding one of the same frame images of the pair, extracting, from an image of a salient part extracted from each of the images in each of the image pyramids, a group of pixels each of whose pixel value is larger than a corresponding threshold value, the threshold value being set for each of the pixels depending on brightness information in a vicinity of a target pixel, generating a map by integrating the extracted group of pixels, binarizing and then integrating the map generated for each of the images in each of the image pyramids which correspond to the same frame images of the pair, and thereby generating, for each of the plurality of frame images, an image feature map representing an estimated distribution of object likelihood on a corresponding one of the plurality of frame images; and
detecting an object on a basis of the image feature map generated.
2. The object detection device according to claim 1 , the processes further comprising:
calculating optical flows between the plurality of frame images captured successively by the camera; and
calculating reliability of the optical flows calculated,
wherein object detection based on at least one of the optical flows and object detection based on the image feature map are performed depending on the reliability calculated.
3. The object detection device according to claim 2 , wherein, out of regions on each of the plurality of frame images, in a region in which reliability of a first optical flow is higher than a threshold value, object detection based on the first optical flow is performed, and in a region in which reliability of a second optical flow is less than or equal to the threshold value, object detection based on the image feature map is performed, the first and second optical flows being included in the optical flows.
4. The object detection device according to claim 2 , the processes further comprising:
removing, out of the optical flows, an optical flow in a direction along a moving direction of the camera, as an optical flow included in a background region of the object,
wherein, out of regions on each of the plurality of frame images, a region in which a third optical flow in a direction not equivalent to that of the optical flow removed has been obtained is determined, as a region in which the object is present, the third optical flow being included in the optical flows.
5. An object detection method comprising:
acquiring, from a plurality of frame images arranged in a time series in an image sequence successively captured by a camera, a pair of same frame images for each of the plurality of frame images and inverting one of the same frame images, generating image pyramids each of which includes images obtained by gradually reducing a corresponding one of the same frame images of the pair, extracting, from an image of a salient part extracted from each of the images in each of the image pyramids, a group of pixels each of whose pixel value is larger than a corresponding threshold value, the threshold value being set for each of the pixels depending on brightness information in a vicinity of a target pixel, generating a map by integrating the extracted group of pixels, binarizing and then integrating the map generated for each of the images in each of the image pyramids which correspond to the same frame images of the pair, and thereby generating, for each of the plurality of frame images, an image feature map representing an estimated distribution of object likelihood on a corresponding one of the plurality of frame images;
calculating optical flows between the plurality of frame images captured successively by the camera;
calculating reliability of the optical flows calculated; and
performing object detection based on at least one of the optical flows and object detection based on the image feature map depending on the reliability calculated.