Title: Realtime 3D Reconstruction of Crop Canopy to estimate Crop Height using SFM or similar perception approaches

Introduction:

PerPlant is a deep-tech startup with the goal of assisting farmers in transitioning to sustainable agriculture through the use of AI and sensor technology. We develop tractor-mounted smart camera systems that enable farmers to monitor their fields in real-time for pest pressure and analyze plant health levels. This helps farmers gain insights on where and how much to spray chemicals like pesticides and fertilizers, reducing chemical costs and mitigating their negative impacts on the environment, human health, biodiversity, and climate.

Crop height is an important parameter in precision agriculture for crop management, yield estimation, and pest management. Accurate and real-time crop height estimation can assist farmers to make timely decisions for better crop management. Traditional crop height measurements methods, such as manual measurements or satellite imagery, are labor-intensive, time-consuming, and lack spatial resolution. To overcome these limitations, computer vision-based approaches have been proposed, including Structure-from-Motion (SFM) techniques. In this thesis, we propose to use SFM on a smart stereo camera system mounted on a tractor's cabin rooftop to perform real-time 3D reconstruction of the crop canopy and estimate the crop height.

Objectives: The main objective of this thesis is to develop and evaluate a system for realtime 3D reconstruction of crop canopy using SFM or a similar approach on a smart stereo camera system mounted on a tractor's cabin rooftop. The specific objectives are:

1. Develop an SFM algorithm for 3D reconstruction of crop canopy from stereo images.
2. Evaluate the accuracy of the 3D reconstruction and crop height estimation on different crop types and under different lighting and weather conditions. The algorithm should be designed considering computationally expense as one of the prime factors, apart from the accuracy.
3. Integrate and deploy the SW on the smart camera system for realtime crop height estimation during normal farming activities.

Methodology: The proposed system will consist of a smart stereo camera system mounted on a tractor's cabin rooftop, which includes a stereo camera pair, a powerful edge computer for real-time data processing, and a GPS module for georeferencing. The stereo camera system will capture stereo images of the crop canopy while the tractor is moving. The stereo images will be processed by an SFM algorithm to reconstruct the 3D point cloud of the crop canopy. The crop height will be estimated from the 3D point cloud by measuring the distance between the highest point of the crop canopy and the ground. The accuracy of the system will be evaluated on different crop types and under different lighting and weather conditions. The system will be integrated with a tractor for real-time crop height estimation during normal farming activities.

Expected outcomes: The expected outcomes of this research are:

1. Development of a method for 3D reconstruction of crop canopies using SFM.
2. Evaluation of the accuracy and precision of the proposed method in estimating crop height.
3. Demonstration of the potential applications of the proposed method in precision agriculture.

Conclusion: The proposed research will develop a real-time crop height estimation system using 3D reconstruction of crop canopy using the SFM approach on a smart stereo camera system mounted on a tractor. The system will provide accurate and reliable crop height measurements, which will be beneficial for crop management and yield estimation. The research will contribute to the field of precision agriculture by providing a cost-effective and efficient solution for crop height estimation.