The aim of the problem is to detect abnormal behaviors (presence of disease) among chickens on a broiler farm. In broiler farms, birds are kept free (without cages) in a shed. The challenge is the detection and tracking of chickens individually and the detection of abnormal behavior (such as the presence of disease in chickens) using the processing of images obtained from surveillance cameras.
Various parameters could indicate the abnormal behavior or disease in poultry. The two main parameters which should be calculated are the mobility of each chicken and how they are dispersed on the farm floor. It should be noted that the mobility of sick chickens decreases.Also, the poultry should be dispersed homogeneously on the level of the chicken farm. Non-homogeneous dispersion of chickens in representative poultry farms indicates a problem, such as the non-uniformity of the ambient temperature on the farm.
The demands of the problem:
1- Detection of chickens and separation of chickens (each segment should contain one chicken)
2- Counting the chickens (number one from the top left and the last number from the bottom right)
3- Tracking the detected chickens from the previous steps
4- Calculation of the movement of each chicken over time
5- Detection of abnormality in the placement of chickens on the floor of the hall
6- Defining a new parameter for diagnosing sick chickens
7- Develop ideas for improving and increasing the coverage of the image processing system for the current field
Additional notes:
1- The calculation must be real-time and the processing speed must be at least three frames per second. The hardware is a laptop GPU (such as ti 1060) or a central processor with eight real cores and 16 threads or in combination. The minimal use of hardware is considered a positive point.
2- The lighting conditions are stable to a large extent; the images are either during the day or at night (the cameras are night vision) – that is, different scenarios can be adopted for the two modes of night and day.
3- The programming language must be Python and be modular as much as possible for different evaluations.
4- In some articles, methods such as examining the chicken’s skeleton and processes based on deep learning have been used to diagnose diseases in chickens. These articles in the references section will be given a separate positive score to get to know the generality of the introduced problem, the use of innovative methods, or the existing methods.
5- Providing explanations in a PDF file is mandatory. The quality of the submitted file also includes points.
6- In the problem requirements section, the code should be implemented in such a way that it is possible to test each of the 1025 steps with a limited number of frames or single frames.
7- For ease and improvement of accuracy, full-page processing of one frame is not required. 76% of the middle of the image from the width – 84% of the middle of the image from the height is the ROI (region of interest) of the problem.
8- The corresponding dataset is unlabeled by default, but there is no limit for labeling. Also, there is no limit to the proposed method, whether based on deep learning or classical image processing methods.
Deadline for submission of works:
November 2022
2022-11-10
Conference website:
https://iccke.um.ac.ir/2022