<div><div><div> How can we transform an agricultural vehicle into an autonomous weeding robot? A robot that can run autonomously through a vegetable field, classify multiple types of weeds from real-time video feed and then spray specific herbicides based of previously classified weeds. In this research, we answer some of the theoretical and practical challenges regarding the transformation of an agricultural vehicle into an autonomous weeding robot. How can we transform an agricultural vehicle into an autonomous weeding robot? A robot that can run autonomously through a vegetable field, classify multiple types of weeds from real-time video feed and then spray specific herbicides based of previously classified weeds. In this research, we answer some of the theoretical and practical challenges regarding the transformation of an agricultural vehicle into an autonomous weeding robot. How can we transform an agricultural vehicle into an autonomous weeding robot? A robot that can run autonomously through a vegetable field, classify multiple types of weeds from real-time video feed and then spray specific herbicides based of previously classified weeds. In this research, we answer some of the theoretical and practical challenges regarding the transformation of an agricultural vehicle into an autonomous weeding robot. <br></div></div></div><div><br></div><div> First, we propose a solution for real-time crop row detection from autonomous navigation of agricultural vehicle using domain knowledge and unsupervised machine learning based approach. We implement projective transformation to transform camera image plane to an image plane exactly at the top of the crop rows, so that parallel crop rows remain parallel. Then we use color based segmentation to differentiate crop and weed pixels from background. We implement hierarchical density-based spatial clustering of applications with noise (HDBSCAN) clustering algorithm to differentiate between the crop row clusters and weed clusters. <br></div><div><br></div><div> Finally we use Random sample consensus (RANSAC) for robust line fitting through the detected crop row clusters. We test our algorithm against four different well established methods for crop row detection in-terms of processing time and accuracy. Our proposed method, Clustering Algorithm based RObust LIne Fitting (CAROLIF), shows significantly better accuracy compared to three other methods with average intersect over union (IoU) value of 73%. We also test our algorithm on a video taken from an agricultural vehicle at a corn field in Indiana. CAROLIF shows promising results under lighting variation, vibration and unusual crop-weed growth. <br></div><div><br></div><div><div> Then we propose a robust weed classification system based on convolutional neural network (CNN) and novel decision-level evidence-based multi-sensor fusion algorithm. We create a small dataset of three different weeds (Giant ragweed, Pigweed and Cocklebur) commonly available in corn fields. We train three different CNN architectures on our dataset. Based on classification accuracy and inference time, we choose VGG16 with transfer learning architecture for real-time weed classification.</div><div> </div><div> To create a robust and stable weed classification pipeline, a multi-sensor fusion algorithm based on Dempster-Shafer (DS) evidence theory with a novel entropy function is proposed. The proposed novel entropy function is inspired from Shannon and Deng entropy but it shows better results at understanding uncertainties in certain scenarios, compared to Shannon and Deng entropy, under DS framework. Our proposed algorithm has two advantages compared to other sensor fusion algorithms. First, it can be applied to both space and time domain to fuse results from multiple sensors and create more robust results. Secondly, it can detect which sensor is faulty in the sensors array and compensate for the faulty sensor by giving it lower weight at real-time. Our proposed algorithm calculates the evidence distance from each sensor and determines if one sensor agrees or disagrees with another. Then it rewards the sensors which agrees with another according to their information quality which is calculated using our novel entropy function. The proposed algorithm can combine highly conflicting evidences from multiple sensors and overcomes the limitation of original DS combination rule. After testing our algorithm with real and simulation data, it shows better convergence rate, anti-disturbing ability and transition property compared to other methods available from open literature.</div></div><div><br></div><div><div> Finally, we present a fuzzy-logic based approach to measure the confidence</div><div> of the detected object's bounding-box (BB) position from a CNN detector. The CNN detector gives us the position of BB with percentage accuracy of the object inside the BB on each image plane. But how do we know for sure that the position of the BB is correct? When we are detecting an object using multiple cameras, the position of the BB on the camera image plane may appear in different places based on the detection accuracy and the position of the cameras. But in 3D space, the object is at the exact same position for both cameras. We use this relation between the camera image planes to create a fuzzy-fusion system which will calculate the confidence value of detection. Based on the fuzzy-rules and accuracy of BB position, this system gives us confidence values at three different stages (`Low', `OK' and `High'). This proposed system is successful at giving correct confidence score for scenarios where objects are correctly detected, objects are partially detected and objects are incorrectly detected. </div></div>

Nowadays, there is a high demand for computer vision applications in numerous application areas, such as autonomous driving or unmanned aerial vehicles. However, the application areas and scenarios are becoming increasingly complex, and their data requirements are growing. To meet these requirements, it needs increasingly powerful computing systems. FPGA-based heterogeneous systems offer an excellent solution in terms of energy efficiency, flexibility, and performance, especially in the field of computer vision. Due to complex applications and the use of FPGAs in combination with other architectures, efficient programming is becoming increasingly difficult. Thus, developers need a comprehensive framework with efficient automation, good usability, reasonable abstraction, and seamless integration of tools. It should provide an easy entry point, and reduce the effort to learn new concepts, programming languages and tools. Additionally, it needs optimized libraries for the user to focus on developing applications without getting involved with the underlying details. These should be well integrated, easy to use, and cover a wide range of possible use cases. The framework needs efficient algorithms to execute applications on heterogeneous architectures with maximum performance. These algorithms should distribute applications across various nodes with low fragmentation and communication overhead and find a near-optimal solution in a reasonable amount of time. This thesis addresses the research problem of an efficient implementation of object detection applications, their distribution across FPGA-based heterogeneous systems, and methods for automation and integration using toolchains. Within this, the three contributions are the HiFlipVX object detection library, the DECISION framework, and the APARMAP application distribution algorithm. HiFlipVX is an open-source HLS-based FPGA library optimized for performance and resource efficiency. It contains 66 highly parameterizable computer vision functions including neural networks, ideally for design space exploration. It extends the OpenVX standard for feature extraction, which is challenging due to unknown element size at design time. All functions are streaming capable to achieve maximum performance by increasing parallelism and reducing off-chip memory access. It does not require external or vendor libraries, which eases project integration, device coverage, and vendor portability, as shown for Intel. The library consumed on average 0.39% FFs and 0.32% LUTs for a set of image processing functions compared to a vendor library. A HiFlipVX implementation of the AKAZE feature detector computes between 3.56 and 4.13 times more pixels per second than the related work, while its resource consumption is comparable to optimized VHDL designs. Its neural network extension achieved a speedup of 3.23 for an AlexNet layer in comparison to a related work, while consuming 73% less on-chip memory. Furthermore, this thesis proposes an improved feature extraction implementation that achieves a repeatability of 72.57% when weighting complex cases, while the next best algorithm only achieves 62.99 %. DECISION is a framework consisting of two toolchains for the efficient programming of FPGA-based heterogeneous systems. Both integrate HiFlipVX and use a joint OpenVXbased frontend to implement computer vision applications. It abstracts the underlying hardware and algorithm details while covering a wide range of architectures and applications. The first toolchain targets x86-based systems consisting of CPUs, GPUs, and FPGAs using OpenCL (Open Computing Language). To create a heterogeneous schedule, it considers device profiles, kernel profiles and estimates, and FPGA dataflow characteristics. It manages synchronization, memory transfers and data coherence at design time. It creates a runtime optimized program which excels by its high parallelism and a low overhead. Additionally, this thesis looks at the integration of OpenCL-based libraries, automatic OpenCL kernel generation, and OpenCL kernel optimization and comparison for different architectures. The second toolchain creates an application specific and adaptive NoC-based architecture. The streaming-optimized architecture enables the reusability of vision functions by multiple applications to improve the resource efficiency while maintaining high performance. For a set of example applications, the resource consumption was more than halved, while its overhead was only 0.015% in terms of performance. APARMAP is an application distribution algorithm for partition-based and mesh-like FPGA topologies. It uses a NoC (Network-on-Chip) as communication infrastructure to connect reconfigurable regions and generate an application-specific hardware architecture. The algorithm uses load balancing techniques to find reasonable solutions within a predictable and scalable amount of time. It optimizes solutions using various heuristics, such as Simulated Annealing and Tabu Search. It uses a multithreaded grid-based approach to prevent threads from calculating the same solution and getting stuck in local minimums. Its constraints and objectives are the FPGA resource utilization, NoC bandwidth consumption, NoC hop count, and execution time of the proposed algorithm. The evaluation showed that the algorithm can deal with heterogeneous and irregular host graph topologies. The algorithm showed a good scalability in terms of computation time for an increasing number of nodes and partitions. It was able to achieve an optimal placement for a set of example graphs up to a size of 196 nodes on host graphs of up to 49 partitions. For a real application with 271 nodes and 441 edges, it was able to achieve a distribution with low resource fragmentation in an average time of 149 ms.