| With the significant improvement of computer performance in data processing, so many computer vision applications have made great development and progress. Video image processing is a very important application research in computer vision, it is also a very important part of the study in the field of image processing.In the applications of multi-targets detection and tracking based on video, in specific and special application scenarios, there are many video processing applications based on different characteristics according to the characteristics of different targets, such as video tracking based on face recognition, video tracking based on contour detection, video tracking algorithm based on color histogram. However, the video has high complexity and uncertainty of moving target itself, there is a great study space and research value of multiple objectives video tracking.In this article’s scenario, the objects to be tracked are farm animals with the same species, such as chicken or ducks in the same region, they have three main features:the same type, no significant differences between their own characteristics; large quantities. Because of the same type, we can’t distinguish animals by different morphological characteristics. Since the characteristic differences between the animals are not obvious, it is not easy to use color, contour or facial features and other characteristics to do identification. Due to a larger number, if we find out the tiny differences as feature to do identification, then the complexity of the algorithm may be higher, when the number is larger, it will seriously affect the recognition speed and accuracy can’t be guaranteed.Therefore, this thesis use the tracking method based on digital identification, in which animals are labeled with a digital sign. Through the identification and tracking of different numbers, we can track animals successfully. The use of digital as a secondary feature to identify objects can reduce the complexity of the animal feature extraction, reducing the amount of calculation, which can improve the recognition efficiency.The aim of applying multi-targets detection and tracking based on video on farm animals is to monitor the health of animals. Animal health status has close contact with its motion. Through the video tracking, monitoring the state of motion of each animal, then doing subsequent analysis of each animal’s amount of exercise and exercise time, we can initially detect possible problematic animals, user can do animal screening timely, reducing the breeding process losses.The main research work and innovations of this thesis are as follows:1. For the same type of multi-targets tracking, a tracking method of using digital as identification feature is proposed. It can effectively recognize the same type of objects with inconspicuous characteristics. And when the number increases, we only need to increase the serial numbers without re-finding feature to modify the algorithm, which is very convenient and efficient.2. For digital identification algorithm, the thesis innovatively proposes the idea of algorithm arbitration, combining with a variety of features for digital identification, arbitrating result based on crossing-line feature, based on the template matching and SVM-based identification, giving the final digital result, reducing the erroneous recognition rate.3. Because digital moves with animal, the direction is random. Therefore, this thesis proposes digital rotation correction method by affine transformation, recognizing digital after direction correction.4. For the problem that numbers may overlap, this thesis innovatively put forward a method that determines digital transition based on historical data, as well as intelligently screens effective data based on the video data range.5. This thesis innovatively proposes the method of dealing data using distributed data processing mechanism, the conventional serial processing of data has been improved by using a parallel way to improve the image processing speed. |
