Home Depot Automated Part Identifier Sensor Team Six

Home Depot Automated Part Identifier Sensor Team Six

Problem Scenario ● Customer goes to Home Depot to repurchase bolts and connectors ● Similarities between parts and large inventory makes identification of parts challenging ● Customers spend significant time searching for part, potentially affecting in-store experience ● Large fraction of Associates time is devoted to helping customers look for parts instead of other tasks

Qualitative Goals ● Build a device to identify bolts and connectors in Home Depot inventory ● Develop user-friendly UI which should be easy to understand for first-time customers ● Improve customer experience by reducing time spent searching for parts ● Reduce time Associate spends searching for parts

Specifications Spec Requirement Battery Last one 16 hour business day Physical Dimensions Less than 20 in x 10 in Processing Time Less than 10 seconds Accuracy Part in top three results > 80%

Design Approach - Hardware ● Image Capture ○ 8 Megapixel camera ○ LED lighting ● Interface Display ○ 7” LCD Touchscreen ● Microcontroller ○ Raspberry Pi will control camera, lighting, and screen, send pictures to the server, and receive the result from the server

Design Approach - Image Processing ● Combine digital signal processing (DSP) approach with machine learning techniques ● Preprocess image using DSP techniques ○ Scaling, Rotation, segmentation ● Identify features that can be used for identification ○ Head type, thread count, length, etc. ● Use probabilistic, categorical identifier to determine part type ○ Display top three results ○ Optional: make recommendations to customer

Classifier Performance ● Classifiers are executed on different data sets than they were trained on, so there must be a function to gauge performance ● Performance should not be gauged based on training data, as this would be biased ● Criterion function for performance assessment is based on accuracy, which involves the confusion matrix

Design Approach - User Interface ● Touch screen LCD user interface ○ Button to initiate processing ○ Top three matches ordered ○ Picture, description, price, part number ● Interface design goals ○ Enjoyable user experience ○ Simple and intuitive graphics

User Interface Draft Layout Please place part on tray and press start. Start Finding your top three matches… Your top three matches are: 1. (picture), (description), (price), (part #) 2. (picture), (description), (price), (part #) 3. (picture), (description), (price), (part #)

Design Approach Home Depot Database Lighting Ima ge D ata ta We LCD Touch Screen User Input Image Data Image Processing ID Result User Input Force Transducer Da ight Micro Controller ID Result Camera Wireless Direct Connection Results Display

Prototype Component Cost Item Cost Raspberry Pi 3 Model B Plus $35. 00 Raspberry Pi 8 Megapixel Camera $29. 95 7” LCD Touchscreen $89. 95 100 g Micro Load Cell $7. 00 8 GB Micro SD Card $6. 79 Raspberry Pi Breakout Board $6. 95 Breadboard $5. 95 Power Supplies/Cables/Wiring/Misc. $20. 00 Packaging/Housing $10. 00 Total $211. 59

Timeline

Responsibilities ● Joseph ○ Assembly of Raspberry Pi circuit ○ Will assist in draft of UI ○ Group Leader - interface between group and advisor/HD ● Christopher ○ Digital signal processing ○ Alignment/segmentation

Responsibilities ● Nicholas ○ Digital signal processing ○ Alignment/segmentation ○ Webmaster ● Matthew ○ Digital signal processing ○ Programming (sending/receiving data) ● Lindsay ○ User Interface design

Status ● Finalizing proposal ● Purchasing parts ● Rotation of image has been tested and completed in MATLAB ● Decided on weekly team meeting time

Status - Image Processing Angle = -47. 3 degrees Matlab software can identify orientation axis and align connector head at top of image

Status - Enclosure Design

Questions?