| Summary: | In a traffic setting, traffic sign detection is a critical task. It is becoming even more important as the number of vehicles on the road grows, increasing the accident rate.
Object detection in embedded devices can be used to alert distracted drivers to traffic signs so that the signs can direct them on the road. Object identification on images can have a high level of accuracy and speed, which is required by a traffic sign detection system to recognize traffic signs quickly and precisely so that road safety can be ensured. However, to achieve the high accuracy and speed requirements, it usually requires a high-performing Graphics Processing Unit (GPU) and considerable internal storage. Unfortunately, most embedded devices do not have high performing GPU and huge internal storage.
Thus, this project proposes the use of efficient and low-bit neural network compression to make the existing object detection model deployable on embedded devices, while ensuring that the accuracy and speed are not too compromised.
|
|---|