Android phone Camera app with Tensorflow

Native android app to support integrated phone camera video analytic processing as limited OCR and its flash light. An external trigger via wi-fi will make the app to start the camera activity and turn on the flash. once on, the camera will take video for app to analyze it.

The target is a moving rectangular metal plate (about 30 cmx 25cm) in front of the camera at a slow speed about 30 CM away. The final picture for analysis will be taken when the front edge of the plate is within a range of distance (user definable) in front of camera, so full plate is visible. Once analysis is successfully complete, the camera and flash will be turned off via the app. Multiple pictures may need to be taken till a good reliable snapshot is taken for successful, accurate result. The system shall rely on flash light and not on ambient light.

The target metal plate has rectangular holes (about 3 cmx 2cm) pattern in it. There are 3 lines with a maximum of 7 holes in each line. Each line may have from 0 to 7 equidistant holes. The holes and space (no holes) define a binary pattern. A hole will be 0 and space will be 1. The algorithm must interpret the hole/space pattern & convert into an equivalent binary pattern. The 3 lines together will give us a total of 21 bits. The numeric equivalent of the 21 bits binary # will be determined and stored in a circular array in a round robbin chronological order. The current index will be used for storing the last read # & for managing on phone display and also comms via wi-fi. Also, keep last n reads in this array and also for display using the current index.

Any kind of errors in the reading will also have to be reported via wi-fi and stored in array. The successful completion and any diagnostics will need to be flagged via wi-fi. The wi-fi comms protocol may be defined by you or we can provide it to you. This must be well defined beforehand.

Provide a teaching/learning mechanism in the algorithm. As the distance of plate from camera, dirt/dust impact and lighting conditions may change. Use of Tensorflow is recommended unless you have a better mousetrap. If yes, get approval. We shall provide sample target plate pictures to selected persons.

The system has to be working with 99.99% accuracy and repeatability with the flash light. The app will be always running as long as power is applied to phone and no log-in should be needed.

Full source code with source of any dependencies (full ownership), A user manual in MS Word and training about the SW code will be provided to us by you.

Optionally, help make a box in 3D printer design to hold the phone and with a clear plastic for Camera/flash and also for the display. This camera system will be kept ON most of the time almost 24x7, so a grommet opening for power will be required in box. The environment is very dusty and dirty so need a mechanism to auto clean (at some frequency) the camera side clear plastic - may be a wiper or something. Rugged HW solutions are required suitable for industrial environment.

We are planning on using a Samsung phone with latest Android OS. If you like to suggest a certain model, we shall consider using it.

Unless otherwise specified by you, the box option is assumed to be included in your bid. Please also submit previous work done only related to ours - not unrelated.

Please read and understand and then respond with YES as the 1st 3 letters of your response as confirmation. Submit questions with your 1st response. Please do NOT copy/paste a pre-set response as a robot.

No upfront payments. The budget is only a placeholder, so ignore it and you need to bid your best bid with best delivery. Any other commercial conditions submit with your response. Any milestones payments (you suggest) will only be released after all source code, supporting docs till that day are submitted and approved by us. If you do not submit milestones with response, will assume one single payment at the end. Please respond only if serious.

Android Tensorflow Google Analytics Android App Development Machine Vision / Video Analytics

ID проекта: #31817457

О проекте