Help with algorithm approach for computer vision

Question

I hope this is the right forum to ask.

I had a client approach me with a demand for a vision system for their assembly line.

The problem they are facing is that the operator sometimes forgets to put all three parts of the product together. They want a vision system that can spot if any of three main components are missing (only missing nothing else).

I have the knowledge to build a covnet in keras but I think that would be a way to big hammer for this nail and I have to go to the client any produce images myself (ergo I can sit there for a day or two and take photos the products). So I might end up with a data set of maybe 200 images. I will try and make some artificially bad images, taking of the different components and photoing that. I don't think I'm allowed to show what the product looks like but I made a sketch. What algorithm can I read up on that would solve this problem in a good and time efficient manner? Demands on the camera and pc for the any suggested algo would also be very much appreciated (resolution etc).

Answer 1

this problem set is in the realm of object detection and image classification,you need have need a good dataset from your client or scrape one yourself,you will need to perform real time object detection in the assembly line so for that case you will need a have a really good convnet that has high accuracy.One way to have such a high accurate model is to perform transfer learning using your preferred models such as inception,caffe,lenet etc. you will need to delve more into learning about anchor boxes,for this kind of problem look at the keras implementation of Yolo algorithm https://towardsdatascience.com/object-detection-using-yolov3-using-keras-80bf35e61ce1