In the forestry industry, the harvester equipment uses a mechanical method for measuring log lengths. Due to several factors, including bark texture, thickness, and wood hardness, this system is prone to inconsistencies and requires frequent recalibration. This thesis investigates the requirements and feasibility of implementing a contactless length measurement system that mitigates these issues.

The system proposed in this thesis uses a single camera to capture images of a log from within a harvester machine. An optical flow estimation method is used to predict the motion of the log between two successive image frames. The mean displacement vector is then calculated using application-specific domain knowledge to improve accuracy.

To facilitate the evaluation of a contactless system, a test platform was designed to implement real-time estimation and collect images to generate a training and validation dataset. Data augmentation techniques were used to introduce application-specific challenges, including motion blur and occlusion. The simulated ground-truth optical flow was produced using a crop and warp method.

Five optical flow estimation methods were compared, looking at their performance in terms of estimated mean motion vector accuracy for image sequences and inferencing time. Dense Inverse Search and SelFlow showed remarkable performance for this application, however, due to SelFlow’s requirement of parallel hardware, Dense Inverse Search was selected to be implemented in real-time on the test platform. The final mean displacement vector computation runs at 90 Hz on the test platform and is within 1.1% of the actual displacement.