Object Detection¶
The usage of artificial intelligence has become a popular technique to apply a “human touch” to technical applications; for example to detect objects, recognize speech, give intuitive answers to questions or simply finding patterns or anomalities in large sets of data. In vehicles, one common task is to have an integrated vision support that is vigilant and more reliable than a human. At CrossControl we are currently focusing on using AI and object detection to find objects in a video stream as a guide for an operator to make better decisions or to increase safety.
In this guide we will try to answer the most common questions about how object detection works and how to use it in our displays.
The following display products are supported:
i.MX8:
CCpilot V700 (using Google Coral USB AI accelerator)
CCpilot V1000 (using Google Coral mPCIe AI accelerator)
CCpilot V1200 (using Google Coral mPCIe AI accelerator)
Glossary¶
| Word/Abbreviation | Explanation |
|---|---|
| AI | Artificial Intelligence – simulation of human intelligence often based on ML |
| Coral | Product family that utilizes the Edge TPU from Google |
| DPS | Detections Per Second |
| Edge TPU | An ASIC "accelerator" that can run inference tasks (Tensor Processing Unit) |
| FPS | Frames Per Second |
| Inference | Conclusion made from facts and/or experiences |
| Inference Engine | Component performing "inference" using AI |
| ML | Machine Learning – the process of making a detection model by using a large set of training data |
| mPCIe | Mini PCI Express standardized expansion bus |
| OpenCV | Open source Computer Vision library |
| Pipeline | A GStreamer pipeline used to play video streams |
| Tensorflow | The Tensorflow open source platform for machine learning |
| Trained Model | A model trained to detect specific information in a data |