Vision Model Optimisation

Service Description

FXC Reference Number: TICH60135 This service focuses on optimising your existing vision models to enhance the speed and accuracy of outputs by leveraging advanced computer vision techniques. Sensor-generated data will be analysed with object detection models such as YOLOv8 with bounding boxes or segmented objects (subject to data annotation) to track object locations in time and locate critical control points. These points may be generated from pre-existing skeletal detection models or custom annotations if available. Machine learning models, including but not limited to Transformers will then analyse the sequence of frames to deliver predictive results. If available, 3D stereoscopic imaging can be included in addition or in place of standard video footage. What to expect: Review and optimisation of object detection and tracking: review current data processing workflow, evaluate performance, and explore possible improvements from standard computer vision libraries (OpenCV) to AI models (e.g. YOLOv8) for object bounding box detection. Improvements could be sought in accuracy, speed or both depending on the case. Segmentation Model Exploration: assess the data potential for training a segmentation model (such as the YOLOv8-?seg family of models) to further narrow object location and potentially improve control point detection. While full annotation of existing data for segmentation might be too time-intensive, transfer learning from pre-trained models will be evaluated on smaller data samples to assess feasibility. A comprehensive evaluation of pre-existing object detection/tracking workflow and any associated control points detection and predictive AI model, with accuracy evaluated on training and testing data across the various object classes. What you takeaway: • Analysis of current procedures to identify optimisation opportunities; • Comparison with state-of-the-art techniques to highlight relevant technical advancements; • Recommendations for enhancing your existing service delivery models. This service is designed for: • Micro-industry; • SMEs. This service must be booked after: • Digital Maturity Assessment; or • Digital Maturity Assessment for Industry.