Abstract: This application provides a target tracking method, including: obtaining a plurality of consecutive picture frames of a target video, and setting a tracked target region of an nth picture frame; performing a feature extraction operation on the tracked target region of the nth picture frame, to obtain a feature map of the tracked target region; calculating a weighted filter corresponding to the input feature map according to a correlation filter algorithm and a mean pooling constraint condition; calculating an output response of an (n+1)th picture frame by using the weighted filter and an input feature map of the (n+1)th picture frame in the plurality of consecutive picture frames, and determining a tracked target region of the (n+1)th picture frame according to the output response of the (n+1)th picture frame, until tracked target regions of all the consecutive picture frames are obtained. This application improves precision and effectiveness of target tracking.

Abstract: Disclosed are methods, apparatuses and systems for CNN network adaption and object online tracking. The CNN network adaption method comprises: transforming a first feature map into a plurality of sub-feature maps, wherein the first feature map is generated by the pre-trained CNN according to a frame of the target video; convolving each of the sub-feature maps with one of a plurality of adaptive convolution kernels, respectively, to output a plurality of second feature maps with improved adaptability; training, frame by frame, the adaptive convolution kernels.

Abstract: In a convolutional neural network (CNN) using an encoder-decoder structure for image segmentation, a multi-scale context aggregation module receives an encoded final-stage feature map from the encoder, and sequentially aggregates multi-scale contexts of this feature map from a global scale to a local scale to strengthen semantic relationships of contexts of different scales to improve segmentation accuracy. The multi-scale contexts are obtained by computing atrous convolution on the feature map for different dilation rates. To reduce computation, a channel-wise feature selection (CFS) module is used in the decoder to merge two input feature maps. Each feature map is processed by a global pooling layer followed by a fully connected layer or a 1×1 convolutional layer to select channels of high activation. By subsequent channel-wise multiplication and elementwise summation, only channels with high activation in both feature maps are preserved and enhanced in the merged feature map.

Abstract: This application provides a target tracking method, including: obtaining a plurality of consecutive picture frames of a target video, and setting a tracked target region of an nth picture frame; performing a feature extraction operation on the tracked target region of the nth picture frame, to obtain a feature map of the tracked target region; calculating a weighted filter corresponding to the input feature map according to a correlation filter algorithm and a mean pooling constraint condition; calculating an output response of an (n+1)th picture frame by using the weighted filter and an input feature map of the (n+1)th picture frame in the plurality of consecutive picture frames, and determining a tracked target region of the (n+1)th picture frame according to the output response of the (n+1)th picture frame, until tracked target regions of all the consecutive picture frames are obtained. This application improves precision and effectiveness of target tracking.

Abstract: In a convolutional neural network (CNN) using an encoder-decoder structure for image segmentation, a multi-scale context aggregation module receives an encoded final-stage feature map from the encoder, and sequentially aggregates multi-scale contexts of this feature map from a global scale to a local scale to strengthen semantic relationships of contexts of different scales to improve segmentation accuracy. The multi-scale contexts are obtained by computing atrous convolution on the feature map for different dilation rates. To reduce computation required by the CNN, a channel-wise feature selection (CFS) module is used in the decoder to merge two input feature maps. Each feature map is processed by a global pooling layer followed by a fully connected layer or a 1×1 convolutional layer to select channels of high activation. By subsequent channel-wise multiplication and elementwise summation, only channels with high activation in both feature maps are preserved and enhanced in the merged feature map.

Abstract: Disclosed are methods for object tracking. In an example, the method comprises: determining a region of interest (ROI) in a first frame of a video sequences; feeding the determined ROI forward through a first CNN (convolutional network) to obtain a plurality of first feature maps in a higher layer of the CNN and a plurality of second feature maps in a lower layer of the first CNN; selecting a plurality of feature maps from the first and second feature maps, respectively; predicting, based on the selected first and second feature maps, two target heat maps indicating a target location for said objects in the current frame, respectively; and estimating, based on the two predicated target heat maps, a final target location for the object in the current frame.

Abstract: Disclosed are methods, apparatuses and systems for CNN network adaption and object online tracking. The CNN network adaption method comprises: transforming a first feature map into a plurality of sub-feature maps, wherein the first feature map is generated by the pre-trained CNN according to a frame of the target video; convolving each of the sub-feature maps with one of a plurality of adaptive convolution kernels, respectively, to output a plurality of second feature maps with improved adaptability; training, frame by frame, the adaptive convolution kernels.

Abstract: Disclosed are methods for object tracking. In an example, the method comprises: determining a region of interest (ROI) in a first frame of a video sequences; feeding the determined ROI forward through a first CNN (convolutional network) to obtain a plurality of first feature maps in a higher layer of the CNN and a plurality of second feature maps in a lower layer of the first CNN; selecting a plurality of feature maps from the first and second feature maps, respectively; predicting, based on the selected first and second feature maps, two target heat maps indicating a target location for said objects in the current frame, respectively; and estimating, based on the two predicated target heat maps, a final target location for the object in the current frame.

Abstract: A visual attention detector includes a feature extraction unit configured to extract a spatiotemporal feature from a local region in a video; a hashing unit configured to convert a spatiotemporal feature value for the local region into a hash value, and to select a training value mapped to the hash using a hash table; and an attention measure determining unit configured to determine an attention measure on the basis of the distance between a spatiotemporal feature value for the local region and the selected training value such that the larger the distance the larger the attention measure.

Abstract: A region-of-interest extraction device is provided with an extraction unit configured to extract one or a plurality of local regions from an input image; a retrieval unit configured to search an image database storing a plurality of images and retrieve an image matching a local region for each of the local regions extracted by the extraction unit; and a relevance score determination unit configured to determine a relevance score for each of the local regions on the basis of the retrieval result from the retrieval unit.

Abstract: A saliency information acquisition device has a local saliency acquisition unit configured to calculate a saliency measure for each pixel in an input image on the basis of information obtained from a local region surrounding each pixel, a candidate-region setting unit configured to set a plurality of candidate regions in the input image, a global saliency acquisition unit configured to calculate a saliency measure for each candidate region in the plurality of candidate regions on the basis of information including a local saliency feature representing an attribute of the saliency measure for each pixel within a candidate region, and a global feature representing an attribute of the candidate regions in relation to the entire input image, and an integration unit configured to combine the saliency measure for each candidate region in the plurality of candidate regions obtained by the global saliency acquisition unit to generate saliency information.

Abstract: A visual attention detector includes a feature extraction unit configured to extract a spatiotemporal feature from a local region in a video; a hashing unit configured to convert a spatiotemporal feature value for the local region into a hash value, and to select a training value mapped to the hash using a hash table; and an attention measure determining unit configured to determine an attention measure on the basis of the distance between a spatiotemporal feature value for the local region and the selected training value such that the larger the distance the larger the attention measure.

Abstract: An image processing apparatus has an image acquisition unit that acquires an image that is to be subjected to processing, a learning sample extraction unit that extracts data of a plurality of learning samples from the image, a classifier learning unit that performs learning of a plurality of classifiers using the plurality of learning samples, a strong classifier generation unit that generates a strong classifier by combining the plurality of learned classifiers, and a saliency map generation unit that generates a saliency map of the image using the strong classifier.

Abstract: A saliency information acquisition device has a local saliency acquisition unit configured to calculate a saliency measure for each pixel in an input image on the basis of information obtained from a local region surrounding each pixel, a candidate-region setting unit configured to set a plurality of candidate regions in the input image, a global saliency acquisition unit configured to calculate a saliency measure for each candidate region in the plurality of candidate regions on the basis of information including a local saliency feature representing an attribute of the saliency measure for each pixel within a candidate region, and a global feature representing an attribute of the candidate regions in relation to the entire input image, and an integration unit configured to combine the saliency measure for each candidate region in the plurality of candidate regions obtained by the global saliency acquisition unit to generate saliency information.

Abstract: A collaborative distance metric leaning method and apparatus for visual tracking utilizes a distance metric to match the target with the best candidate. A collaborative distance metric learning algorithm is used for visual tracking by fusing two distance metrics learnt for different representations of the target. To further improve the performance of the trained distance metrics, a sample selection mechanism is also used that exploits the intrinsic structure of dense sampling.

Abstract: An image processing device and a method performed by a computer, the image processing device comprising: a processing unit configured to operate as an image acquisition unit configured to acquire an image; a similarity map generator configured to calculate, with a region constructed with one or a plurality of pixels in the image as a constituent unit, a first similarity map based on a first algorithm, the first similarity map representing a degree of similarity between the region and a marginal region of the image, and calculate a second similarity map based on a second algorithm, the second similarity map representing a degree of similarity between the region and the marginal region of the image; and a saliency map generator configured to integrate the first similarity map and the second similarity map to generate a saliency map.

Abstract: An image processing apparatus has an image acquisition unit that acquires an image that is to be subjected to processing, a learning sample extraction unit that extracts data of a plurality of learning samples from the image, a classifier learning unit that performs learning of a plurality of classifiers using the plurality of learning samples, a strong classifier generation unit that generates a strong classifier by combining the plurality of learned classifiers, and a saliency map generation unit that generates a saliency map of the image using the strong classifier.

Abstract: An event detection apparatus determines the occurrence of an abnormal event based on input data without predefining and learning normal or abnormal patterns. A first data obtaining unit obtains first data. A feature quantity classifying unit obtains a feature quantity corresponding to the first data, generates a plurality of clusters for classifying the obtained feature quantity, and classifies the feature quantity into a corresponding one of the clusters. A learning unit learns a plurality of identifiers using feature quantities classified into the clusters. A second data obtaining unit obtains second data. An identifier unit inputs a feature quantity corresponding to the second data into the plurality of learned identifiers, and receives an identification result from each identifier. A determination unit determines whether the second data includes an identification target event based on the obtained identification result.

Abstract: An image processing device and a method performed by a computer, the image processing device comprising: a processing unit configured to operate as an image acquisition unit configured to acquire an image; a similarity map generator configured to calculate, with a region constructed with one or a plurality of pixels in the image as a constituent unit, a first similarity map based on a first algorithm, the first similarity map representing a degree of similarity between the region and a marginal region of the image, and calculate a second similarity map based on a second algorithm, the second similarity map representing a degree of similarity between the region and the marginal region of the image; and a saliency map generator configured to integrate the first similarity map and the second similarity map to generate a saliency map.