Autonomous systems for animal observation that can reach and even surpass human accuracy, based on viso.ai. Animal monitoring with computer vision offers the possibility of non-intrusive animal monitoring in livestock farming.
Livestock products play a vital role in satisfying global food requirements. The demand is projected to increase by 70 percent, especially due to a growing population estimated to reach 9.6 billion by 2050. While the demand is growing, the need for new technologies such as deep learning in the agriculture market is bigger than ever.
Especially the growing demand for animal monitoring systems make visual Artificial Intelligence (AI) to become one of the most promising technologies to automate inspection and cut costs to sustain under high cost pressure and fierce competition. In fact, the livestock monitoring market is expected to grow with a CAGR of 18.2% to reach a total market size of USD 13.3 billion by 2027. This is driven by factors such as animal comfort management, reproduction management or the early detection of infectious diseases.
Early Detection of Infectious Diseases
Recently, outbreaks of infectious diseases (livestock diseases etc.) have been covered widely in newspapers. The most common example is the African Swine Fever with devastating death tolls, millions of animals getting culled, and finally resulting in significant economic loss.
Early detection of infectious diseases will help to greatly reduce the spread of bacteria, viruses, or parasites. Early symptoms include fever and weakness. These are not trivial to be detected via simple observation, especially during an early stage of infection.
Moreover, given the predominantly low staff-per-animal ratios, the short periods of time available for observation only permit the detection of substantial changes, which may be too late for effective intervention in most cases.
Computer Vision for Animal Monitoring
Given the fast-paced advances in Machine Learning, it has become viable to develop autonomous systems for animal monitoring and observation with computer vision that can reach and even surpass human accuracy.
Focusing on the current outbreak of the African Swine Fever, there are studies showing that video processing offers the possibility of non-intrusive animal monitoring in real time. The results show that after an infection, a significant decrease in animal motion can be detected at just four days after experimental infection with the ASF virus. This was the same day that the virus was detected in blood using qPCR.
However, the methodology used in this example was tested on only eight animals. Additionally, it could not exclude human-related motion and is inaccurate at tracking the animals individually. Moreover, a scalable system would be needed to cover the entire farm.
The Results of our Animal Monitoring PoC
Animals infected with the ASF virus show a progressive deceleration in performing daily activities. This is caused by muscle weakness from early stages of infection. Through continuous monitoring of animal behavior with Deep Learning techniques, animal slowdown and weakening can be detected. Viso.ai has developed a custom Deep Learning model for this use case that allows to be used at-scale:
- Comprehensive dataset of more than 100’000 swines of different age and size
- Optimized for covering large areas (fish-eye cameras)
- Detect and compare activity levels at >96% accuracy
- Detect, split and track grouped animals with Computer Vision
- Real-time anomaly notification via API to third party systems
The Deep Learning inference is running on the Edge using modern AI hardware (NVIDIA Jetson TX 2) with the software platform Viso Suite. Real-time action triggers are configured based on the individual use case to detect various behavioral changes.
Through real-time video processing, Computer Vision can surpass human accuracy. It enables new ways of autonomous, non-intrusive systems for animal monitoring and inspection.
If you enjoyed reading this article, you might be interested in:
- Read about Object Detection, the technology behind this use case
- See complete list of real-world Computer Vision applications
- Read about reasons why Computer Vision projects fail
- Learn how to get started Computer Vision projects