Overview data requirement specification for visual inspection

--- title: Overview data requirement specification for visual inspection tags: COTAI Projects --- # Table of Contents [toc] # Data annotation overview **Sample data**. Sample images for the following * Sample images for overall process * Drawing of cables (as given in the PDF file) * *Expected number of images*. $10$ to $20$ * Sample images about several cable types (5 different cable types) * *Expected number of images*. $1$ images per type * More sample images like images given in each step of the PDF * *Expected number of images*. $10$ to $20$ * A sample video recording the full checking procedure, i.e. including all checking steps >**NOTE**. The procedure should be captured by the camera, which is supposed to be used in real checking >**NOTE**. The first 10 seconds of the video should be the background only (for background removal purpose) * Sample images for step 1 in **Overview plan for visual inspection**, i.e. must cover the following cases * When the cable has some abnormal point (5 images) * When the cable has no abnormal point (5 images) * Sample images for step 2 in **Overview plan for visual inspection**, i.e. must cover the following cases * When some crimps are not connected (5 images) * When all crimps are connected (5 images) * Sample images for step 3 in **Overview plan for visual inspection**, i.e. must cover the following cases * When some connection points are not connected (5 images) * When all connection points are connected (5 images) * Sample images for step 4 in **Overview plan for visual inspection**, i.e. must cover the following cases * When the tubing is damaged (5 images) * When the tubing is not damaged (5 images) * When the tubing does not conform the shape of the cable (5 images) * When the tubing conforms the shape of the cable (5 images) * When the tubing is not located correctly (5 images) * When the tubing is located correctly (5 images) * Sample images for step 5 in **Overview plan for visual inspection**, i.e. must cover the following cases * When the label is not located correctly (5 images) * When the label is located correctly (5 images) * When the label is not oriented correctly (5 images) * When the label is oriented correctly (5 images) * Sample images for step 6 in **Overview plan for visual inspection**, i.e. * The image capturing the cable rolling about some firm frame (5 images) **Data for training and evaluating deep learning models**. * Anchor images of rolled cable (see **Flow for step 1** in **Overview plan for visual inspection**) for overall anomaly detection * *Brief*. To ensure best performance, collected anchor images should capture all parts of the cable, i.e. no part is hidden or coverred * *Expected number of images*. $10$ images for each view (will request more if needed) >**NOTE**. The types of view should be defined clearly >**NOTE**. Need more investigation to define the views for capturing anchor images * Data for modular connector detection (i.e. images with connector bounding boxes annotated) * *Brief*. According to the PDF file, there are 3 different views of the connector involved in the checking steps $\to$ We need data for all of these views (and other views involved in the checking process, if any) * *Expected number of images*. About $1,500$ images (with annotations) for each view (will request more if needed) * Data for connection point detection (i.e. squares in the connector) (check **Flow for step 3** in **Overview plan for visual inspection** for more details) * *Brief*. To ensure best performance, we prefer the annotations of the squares to be tetragons, rather than bounding boxes $\to$ Just annotate 4 corners of each square * *Expected number of images*. About $2,000$ images (with annotations) (will request more if needed) >**NOTE**. The order, according to which the squares are annotated, should be defined clearly and followed strictly >**NOTE**. The order, according to which the corners of each square are annotated, should be defined clearly and followed strictly * Anchor images of rolled cable (see **Flow for step 4** in **Overview plan for visual inspection**) for tubing anomaly detection * *Brief*. To ensure best performance, collected anchor images should capture all parts of the tubing, i.e. no part is hidden or coverred * *Expected number of images*. $10$ images for each view (will request more if needed) >**NOTE**. The types of view should be defined clearly >**NOTE**. Need more investigation to define the views for capturing anchor images # Question * Is the tubing look the same for all types of cable? * Should we detect the crimps by YOLOv5 or classical CV?