The repo contains the implementation of a motion detection device that detects the up-down, right-left, circular and idle motion of the target machine i.e. mobile phone or any supported board
Connect your phone to the Edge Impulse platform
once your phone is connected start collecting by clicking starting recording on your phone
Now shake your phone right-left 10 to 20 times and do the same for up-down and circular motion. For idleness leave your phone and on a plane desk and then record. Your data will appear in the Data Acquisition section don't forget to label your data.
Head to the Impulse design page in your project. Add a Spectral Analysis processing block. Add a Classification-BrainChip Akida and a K-means Anomaly Detection block to the learning blocks section.
After saving impulse head on to the spectral features and generate features.
Head to the Classification-BrainChip Akida page in your project. Click Start Training. After a few minutes, the model should be done training. Now you can monitor the accuracy of the Model
Now go to the Anomaly detection page and select the features to train. I have selected RMS values of acceleration along the x, y, and z axis.Go to the Deployment Page and Scan the QR code using your phone.
The model predicts the probability of the state of the device. If the device is idle the probability of idle will be one and same goes for the up-down, right-left and circular motion.