Deep Learning
Judgment System to Detect Truck Turning Right without Stopping
Shaanxi, China · Mar. 2022 - Present
National Innovation and Entrepreneurship Training Project for University(China)
Achieved Monocular 3D detection based on M3D-RPN machine learning model, judged the vehicle driving direction and speed by the movement and inter-frame offset
Achieved results: 87% recognition accuracy at a real-time speed of ~34 FPS on Tesla V100
Shaanxi, China · Sep 2022. - Dec 2022.
Personal Project
Based on the YOLOv4 and paired with the MobilenetV3 to lightweight the model
Used the K-means to construct an Anchor box for the traffic lights dataset
Achieved results: 93.85% mAP at a real-time speed of ~56 FPS on Tesla V100, which is higher than the original YOLOv4 result: 93.59% mAP at a speed of ~47 FPS
| mAP | FPS | |
|---|---|---|
| YOLOv4 | 93.59% | 47 |
| YOLOv4-MobileNet | 92.33% | 59 |
| YOLOv4-MobileNet (generate anchor) | 94.38% | 56 |