Client is a leading global manufacturer of portable stud detectors, a class of devices to detect the presence of a frame behind wall surfaces. As a stud detector is moved across a detection zone, the sensors in the device have to predict with reasonable accuracy, the presence of a frame behind the wall surface. Unfortunately, due to calibration errors which fail to capture real life variations and environmental factors, accuracy of the detector was poor. Client wanted to explore the feasibility of using machine learning algorithms to enhance the stud detectors accuracy.
INXITE OUT APPROACH
Tracking stud detector location in real time
First component involved creation of a heat map of the wall based on the stud detectors readings. As the stud detector is moved across the detection zone, it is continuously imaged and image processing algorithms were used to capture the location of the stud detector with respect to a reference point in the wall. Additionally, it also captures the sensor reading at each such location, thus creating the heat map of the wall. Such heat maps (figure below) essentially are the raw data on which the machine learning models would be trained and tested.
Detecting studs based on stud detector readings
Generated heat maps and corresponding tagged walls with hidden frames were used to train supervised machine learning models. Such machine learning models are essentially classification models which take the sensor readings of a particular section of the wall and predict the presence of a frame behind the wall. Models developed were applied on further test heats maps to assess their accuracy. The output of the generated model is illustrated in the figure below.
Machine learning trained models generated prediction with over 90% accuracy, a significant improvement over the existing detection algorithms