This thesis focuses on the creation of a shape detection algorithm designed to work with 3D point cloud data, which is essential for many industrial automation tasks. The algorithm was developed using traditional image processing techniques and was tested on a variety of shapes, ranging from simple forms like spheres and cones to more complex objects such as dice or cubes. The goal was to create an efficient and accurate algorithm that can detect these shapes in real-time, which is crucial in industries where speed and precision are important.To ensure the captured data was accurate, the Photoneo 3D scanner was carefully calibrated. This scanner was used to generate point cloud data, which is essentially a digital representation of the surfaces of objects. The data was then processed by the algorithm, which identifies and refines the shapes it detects. The results of the testing showed that the algorithm is particularly good at detecting simple geometric shapes like spheres, where it performed with high accuracy. However, when it came to more complex shapes like dice, the accuracy dropped slightly. This drop was more noticeable when multiple objects were present in the data, as the algorithm struggled to distinguish between closely positioned objects.In terms of performance, the algorithm was efficient, especially when handling simple shapes. It could process the data quickly enough to be used in real-time applications. However, more complex shapes took longer to process, and the algorithm required more memory to handle these cases.

The study also considered how the algorithm compares to other methods, such as traditional image processing and deep learning approaches. While deep learning techniques tend to offer higher accuracy, they also require much more computational power and memory, making them less practical for real-time applications in some cases. On the other hand, traditional methods are faster and require less memory but may not handle complex shapes as effectively. The algorithm developed in this thesis strikes a balance between these approaches, offering good performance with moderate computational demands.Additionally, the study aimed to incorporate the Volume of Interest (VOI) algorithm into the detection process. However, it was found that the VOI algorithm did not perform as well as expected, particularly due to its sensitivity to noise in the data. As a result, the outcomes from the VOI algorithm were not included in the final analysis, indicating an area for future improvement. The thesis concludes that the developed algorithm is effective for most scenarios but could benefit from further improvements, such as incorporating deep learning techniques to handle more complex environments. This work has practical implications for industries that require accurate and fast shape detection, such as in quality control or automated assembly lines. Future research could focus on optimizing the algorithm further, testing it with more varied and complex shapes, and exploring how it can be integrated with other technologies for even better performance.