Object detection method and object detection apparatus

1. An object detection method comprising:

mapping at least one image frame in an image sequence into a three dimensional physical space, to obtain three dimensional coordinates of each pixel in the at least one image frame;

extracting a foreground region in the at least one image frame;

segmenting the foreground region into a set of blobs;

detecting, for each blob in the set of blobs, an object in the blob through a neural network based on the three dimensional coordinates of at least one predetermined reference point in the blob to obtain an object detection result; and

matching the object detection results of at least two image frames to obtain an object tracking result;

wherein matching of the object detection results of the at least two image frames comprises:

extracting feature information of each object detected in each image frame of the at least two image frames, wherein the feature information comprises at least one of location information, appearance information, and texture information of the object;

for each object in an image frame of the at least two image frames, assigning a plurality of weights to a plurality of links between the object and each object in other image frames of the at least two image frames respectively based on the feature information, wherein each of the weights indicates a probability of the object being matched with each object in the other image frames; and

matching the at least two image frames in the image sequence based on the weights to obtain the object tracking result;

calculating Euclidean distances between the object and each object in other image frames of the at least two image frames; and

assigning the weights to the links between the object and each object in other image frames of the at least two image frames respectively based on the Euclidean distances;

wherein the smaller the Euclidean distance is, the larger the weight assigned to the link is.

2. The object detection method of claim 1 , wherein obtaining of the three dimensional coordinates of each pixel in the image frame comprises:

determining at least three non-collinear pixels on a predetermined reference plane in an image space;

acquiring deep information of the at least three pixels;

establishing a mapping transformation from the image space to the three dimensional physical space based on the deep information of the at least three pixels; and

mapping each pixel in the image frame to the three dimensional physical space through the mapping transformation to obtain the three dimensional coordinates of each pixel in the image frame.

3. An object detection apparatus comprising:

a three dimensional coordinate obtaining unit operative to map at least one image frame in an image sequence into a three dimensional physical space, to obtain three dimensional coordinates of each pixel in the at least one image frame;

a foreground extracting unit operative to extract a foreground region in the at least one image frame;

a segmenting unit operative to segment the foreground region into a set of blobs;

a detecting unit operative to detect, for each blob in the set of blobs, an object in the blob through a neural network based on the three dimensional coordinates of at least one predetermined reference point in the blob to obtain an object detection result; and

a tracking unit operative to match the object detection results of at least two image frames to obtain an object tracking result;

wherein the tracking unit comprises:

a feature extracting unit operative to extract feature information of each object detected in each image frame of the at least two image frames, wherein the feature information comprises at least one of location information, appearance information, and texture information of the object;

a weight assigning unit operative to assign, for each object in an image frame of the at least two image frames, a plurality of weights to a plurality of links between the object and each object in other image frames of the at least two image frames respectively based on the feature information, and wherein each of the weights indicates a probability of the object being matched with each object in the other image frames; and

a matching unit operative to match the at least two image frames in the image sequence based on the weights to obtain the object tracking result;

the weight assigning unit operative to calculate Euclidean distances between the object and each object in other image frames of the at least two image frames, and assign the weights to the links between the object and each object in other image frames of the at least two image frames respectively based on the Euclidean distances, during assigning the plurality of weights to the plurality of links between the object and each object in other image frames of the at least two image frames;

the smaller the Euclidean distance is, the larger the weight assigned to the link is.

4. The object detection apparatus of claim 3 , wherein the three dimensional coordinate obtaining unit comprises:

a pixel determining unit operative to determine at least three non-collinear pixels on a predetermined reference plane in an image space;

a deep information acquiring unit operative to acquire deep information of the at least three pixels;

a mapping transformation establishing unit operative to establish a mapping transformation from the image space to the three dimensional physical space based on the deep information of the at least three pixels; and

a mapping unit operative to map each pixel in the image frame to the three dimensional physical space through the mapping transformation to obtain the three dimensional coordinates of each pixel in the image frame.

5. An object detection apparatus comprising:

a processor;

a memory comprising computer program instruction, wherein the computer program instructions operative to, when read by the processor, cause the object detection apparatus to perform at least the following:

mapping at least one image frame in an image sequence into a three dimensional physical space, to obtain three dimensional coordinates of each pixel in the at least one image frame;

extracting a foreground region in the at least one image frame;

segmenting the foreground region into a set of blobs;

detecting, for each blob in the set of blobs, an object in the blob through a neural network based on the three dimensional coordinates of at least one predetermined reference point in the blob to obtain an object detection result; and

matching the object detection results of at least two image frames to obtain an object tracking result;

wherein matching of the object detection results of the at least two image frames, performed by the object detection apparatus caused by the computer program instructions read by the processor, comprises:

extracting feature information of each object detected in each image frame of the at least two image frames, wherein the feature information comprises at least one of location information, appearance information, and texture information of the object;

for each object in an image frame of the at least two image frames, assigning a plurality of weights to a plurality of links between the object and each object in other image frames of the at least two image frames respectively based on the feature information, wherein each of the weights indicates a probability of the object being matched with each object in the other image frames; and

matching the at least two image frames in the image sequence based on the weights to obtain the object tracking result;

calculating Euclidean distances between the object and each object in other image frames of the at least two image frames; and

assigning the weights to the links between the object and each object in other image frames of the at least two image frames respectively based on the Euclidean distances;

wherein the smaller the Euclidean distance is, the larger the weight assigned to the link is.

6. The object detection apparatus of claim 5 , wherein obtaining of the three dimensional coordinates of each pixel in the image frame, performed by the object detection apparatus caused by the computer program instructions read by the processor, comprises:

determining at least three non-collinear pixels on a predetermined reference plane in an image space;

acquiring deep information of the at least three pixels;

establishing a mapping transformation from the image space to the three dimensional physical space, based on the deep information of the at least three pixels; and

mapping each pixel in the image frame to the three dimensional physical space through the mapping transformation to obtain the three dimensional coordinates of each pixel in the image frame.