Periodic Reporting for period 2 - BIRD RELEASE (BIRD RELEASE: REpelLEnt Auto-SystEm)
Periodo di rendicontazione: 2018-06-01 al 2019-11-30
For decades, farmers have been in conflict with birds. Changing national agricultural regimes and EU directives are among the main reasons for a growing geese population: 60% increase since 1993 to 5 million. Throughout Northern Europe, the rising geese population causes great damage to farmers’ crops (up to 68k per farmer per incident). Bird Control Group offers solutions which eradicate this damage without harming wildlife or impacting geese populations. The overall objective of the project is to bring our next generation system to market: the Autonomous Bird Control Deterrent System (ABCD). This system is able to successfully scare geese without causing harm and has the potential to revolutionise the way farmers deal with geese problems. When a goose enters an agricultural field, it is immediately detected by a smart software-powered camera system that is continuously scanning the field. Based on the movement characteristics of geese, a pattern is chosen to aim a diode laser towards the birds, causing all birds to scare off and leave immediately. Our field trials and scientific research have proven that lasers are 96% effective in deterring birds when applied correctly. Research has shown that our laser is animal-friendly (acknowledged by the World Wildlife Fund) and that birds do not become accustomed to laser, thus giving it a long term effect. ABCD is based on 4 years of R&D and is capable of deterring more than 80% of birds from farmland, while harming none. As a result, farmers are able to avoid thousands of euros in yearly damage.
Objective 1: Success in component optimisation.
Objective 2: Success in assessment of performance and certification.
Objective 3: Success in commercialisation strategy (demonstration) and final business plan development.
Objective 1: Success in component optimisation.
Objective 2: Success in assessment of performance and certification.
Objective 3: Success in commercialisation strategy (demonstration) and final business plan development.
WP1 - Project Management
WP2 - Design & hardware development
- High level design finished.
- Third-party hardware tested.
- Most viable alternatives selected.
- Diode laser designed. First prototype produced. First prototype tested. Test results communicated to CNI. Redesign applied. First prototype series produced. First prototype series tested. Test results communicated to CNI. Redesign applied. First pilot batch ordered. First pilot batch received. First pilot batch tested. Test results documented.
- Develop pan tilt unit motor in collaboration with TCITS (LonTrend)
- Most viable off-the-shelf PTU identified. Redesign of PTU performed. First prototype produced. First prototype tested. Test results communicated to LonTrend. Redesign applied. Second prototype produced. Second prototype tested. Test results communicated to LonTrend. Redesign applied. Third prototype tested. Test results communicated to LonTrend. Redesign applied. First pilot batch ordered.
WP3 - Software development
- BCG has selected and compared numerous unsupervised learning algorithms that perform background subtraction. They have been programmed in C++ and evaluated on a dataset of videos to qualitatively and quantitatively assess the performance. BCG has selected an algorithm that works best for their purpose and tuned its parameters to get the best performance. This algorithm was then integrated in our software and further optimized by investigating parallelization on the CPU and GPU. The latter turned out to greatly improve the performance. As such, the selected algorithm was implemented in C++ using Open CV and CUDA to run it on the GPU and achieve optimal performance.
- BCG has generated a database of birds and non-birds by scraping from the web. In a literature study, BCG has investigated to use either Haar transformations, Local Binary Patterns (LBP) or Histogram of Oriented Gradient (HOG) features for their algorithm. The latter was selected. Similarly, BCG investigated whether or not to use Boosting (such as AdaBoost) or machine learning algorithms such as Support Vector Machines (SVM). The latter was selected. BCG has systematically varied all the parameters of the HOG features and the SVM algorithm. For every set of parameters BCG trained the algorithm and verified the result with the ground truth data. This sequence was repeated until the best performance was obtained. This gave BCG an idea about the best set of parameters to use. This was tested in Python and implemented in C++ and integrated into their system. It was tested in the field to obtained valuable log data.
- Different tracking algorithms were selected and compared via a qualitative and quantitative analysis. The Kernelized Correlation Filter (KCF) was selected as the best possible option for birds that are relatively close.
- The laser module saves the time it has been powered on and the PTU modules saves the distance it has moved. Both values can be requested using the communication protocol. BCG is initiating duration tests to investigate how long the laser module and PTU module can successfully operate. BCG is investigating how the laser brightness affects its life.
- BCG has integrated a camera from the detection module into the system and implemented patterns that move the PTU to look around for 360 degrees.
WP4 - Testing, validating, certificating, demonstrating
- BCG had internal sessions and has a proposal from Wageningen.
- BCG had multiple internal meetings on IPR and certicifation and is now working on draft patent texts. The first draft patent text is sent to the patent lawyer of VO.
- BCG has demonstrated parts of the device to customers.
WP2 - Design & hardware development
- High level design finished.
- Third-party hardware tested.
- Most viable alternatives selected.
- Diode laser designed. First prototype produced. First prototype tested. Test results communicated to CNI. Redesign applied. First prototype series produced. First prototype series tested. Test results communicated to CNI. Redesign applied. First pilot batch ordered. First pilot batch received. First pilot batch tested. Test results documented.
- Develop pan tilt unit motor in collaboration with TCITS (LonTrend)
- Most viable off-the-shelf PTU identified. Redesign of PTU performed. First prototype produced. First prototype tested. Test results communicated to LonTrend. Redesign applied. Second prototype produced. Second prototype tested. Test results communicated to LonTrend. Redesign applied. Third prototype tested. Test results communicated to LonTrend. Redesign applied. First pilot batch ordered.
WP3 - Software development
- BCG has selected and compared numerous unsupervised learning algorithms that perform background subtraction. They have been programmed in C++ and evaluated on a dataset of videos to qualitatively and quantitatively assess the performance. BCG has selected an algorithm that works best for their purpose and tuned its parameters to get the best performance. This algorithm was then integrated in our software and further optimized by investigating parallelization on the CPU and GPU. The latter turned out to greatly improve the performance. As such, the selected algorithm was implemented in C++ using Open CV and CUDA to run it on the GPU and achieve optimal performance.
- BCG has generated a database of birds and non-birds by scraping from the web. In a literature study, BCG has investigated to use either Haar transformations, Local Binary Patterns (LBP) or Histogram of Oriented Gradient (HOG) features for their algorithm. The latter was selected. Similarly, BCG investigated whether or not to use Boosting (such as AdaBoost) or machine learning algorithms such as Support Vector Machines (SVM). The latter was selected. BCG has systematically varied all the parameters of the HOG features and the SVM algorithm. For every set of parameters BCG trained the algorithm and verified the result with the ground truth data. This sequence was repeated until the best performance was obtained. This gave BCG an idea about the best set of parameters to use. This was tested in Python and implemented in C++ and integrated into their system. It was tested in the field to obtained valuable log data.
- Different tracking algorithms were selected and compared via a qualitative and quantitative analysis. The Kernelized Correlation Filter (KCF) was selected as the best possible option for birds that are relatively close.
- The laser module saves the time it has been powered on and the PTU modules saves the distance it has moved. Both values can be requested using the communication protocol. BCG is initiating duration tests to investigate how long the laser module and PTU module can successfully operate. BCG is investigating how the laser brightness affects its life.
- BCG has integrated a camera from the detection module into the system and implemented patterns that move the PTU to look around for 360 degrees.
WP4 - Testing, validating, certificating, demonstrating
- BCG had internal sessions and has a proposal from Wageningen.
- BCG had multiple internal meetings on IPR and certicifation and is now working on draft patent texts. The first draft patent text is sent to the patent lawyer of VO.
- BCG has demonstrated parts of the device to customers.
Expected results:
1. Successful autonomous bird detection combined with machine learning. Offers an easy to build system for production scale-up.
2. A successful demonstration accompanied with business case for PR purposes.
3. Validated and effective roadmap to sell ABCD.
Potential impacts of an efficient solution for farmers:
1. Provide an effective and long term approach to scare birds.
2. Repel birds in an animal friendly manner.
3. Do this without causing nuisance or pollution to the environment.
4. Can be incorporated in Integrated Pest Management farming.
5. Deliver a cost efficient product which earns back its investment while being used.
6. Combine these into an automated solution that requires minimal effort from users.
7. Give farmers insight in the effectiveness of their repellent method.
Socio-economic impact and the wider societal implications:
1. Advancing innovations in Integrated Pest Management (IPM). This is a broad-based approach that integrates practices for the economic control of pests. IPM aims to suppress pest populations below the economic injury level. ABCD’s emphasis is on control, not on eradication of pests, we believe mechanical controls should be used before biological and chemical controls. Our solution does just that, with success.
2. Resource-efficient eco-innovative food production and processing. The ABCD system is resource-efficient and sustainable. While improving profitability for farms, wildlife and nature are not compromised as the system does not harm birds. Also, resources are used more efficiently as less of the crops are damaged by birds. Most-importantly, ABCD replaces incumbent chemical pest control methods. Thus food production occurs in a resource-efficient and eco-innovative way.
3. Reduction of food losses and waste on farm and along the value-chain. With the use of the ABCD system, less crops are being eaten than in comparison with other bird deterrent systems. This creates a higher yield for farmers and across the whole value-chain.
1. Successful autonomous bird detection combined with machine learning. Offers an easy to build system for production scale-up.
2. A successful demonstration accompanied with business case for PR purposes.
3. Validated and effective roadmap to sell ABCD.
Potential impacts of an efficient solution for farmers:
1. Provide an effective and long term approach to scare birds.
2. Repel birds in an animal friendly manner.
3. Do this without causing nuisance or pollution to the environment.
4. Can be incorporated in Integrated Pest Management farming.
5. Deliver a cost efficient product which earns back its investment while being used.
6. Combine these into an automated solution that requires minimal effort from users.
7. Give farmers insight in the effectiveness of their repellent method.
Socio-economic impact and the wider societal implications:
1. Advancing innovations in Integrated Pest Management (IPM). This is a broad-based approach that integrates practices for the economic control of pests. IPM aims to suppress pest populations below the economic injury level. ABCD’s emphasis is on control, not on eradication of pests, we believe mechanical controls should be used before biological and chemical controls. Our solution does just that, with success.
2. Resource-efficient eco-innovative food production and processing. The ABCD system is resource-efficient and sustainable. While improving profitability for farms, wildlife and nature are not compromised as the system does not harm birds. Also, resources are used more efficiently as less of the crops are damaged by birds. Most-importantly, ABCD replaces incumbent chemical pest control methods. Thus food production occurs in a resource-efficient and eco-innovative way.
3. Reduction of food losses and waste on farm and along the value-chain. With the use of the ABCD system, less crops are being eaten than in comparison with other bird deterrent systems. This creates a higher yield for farmers and across the whole value-chain.