交通标志识别项目 putsi/tsaraisa · GitHub

问：识别交通标志，怎么确定图片里有交通标志，具体在哪里？答：这个应该算目标识别和物体识别，CV的经典教程都有object recognition的内容可以参考。具体到交通标志识别，Github上有些开源代码 Matlab C++ Java的都有

Tsaraisa was run with "./tsaraisa.py -s -g -c lbpCascade.xml -M" on demo screenshot.

What does it do?

• Detect traffic signs.
• Recognize speed limits in signs.
• (optional) Compare GPS-speed to speed limit.
• (optional) Run user command when overspeeding.

How?

GPS-class

• Uses threading to update GPS-info automatically.
• Gets speed from GPS-daemon using python bindings.

Frame handling

• Reads frame from webcam.
• Converts frame to grayscale with cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY).
• (optional) Morphological Open/close.
• (optional) Equalizes histograms with cv2.equalizeHist().
• Downscale frame with multiplier n.

Traffic sign detection

• Uses LBP- or HAAR-cascade to detect speed signs.
• LBP-cascade was trained with 2000 positive and 1000 negative images.
  • Negative image is image of road without speed limit sign.
  • Positive image is image of road with speed limit sign.
• LBP- and HAAR-detection allows great differences in lightning.
• LBP- and HAAR-detection works great on low performance machines.

Recognizing speed limits

• Uses Fast Approximate Nearest Neighbor Search Library feature matching.
• Creates "keypoints" of detected sign and compares them to all known speed limits (files in data-folder).
• Keypoint match distances need to be inside a threshold.
• Match with biggest proper keypoint amount will be returned.
• It is fast and pretty accurate with different lightning conditions.

Speed assist

• When new speed limit is detected it is added as current speed limit.
• After every frame script compares current speed to current speed limit.
• Script runs specified command when overspeeding (e.g. "beep").

Requirements

Required software

• OpenCV >=3.0.0
• Python >=2.7.3
• LibAV >=0.8.10
• (optional) gpsd && python-gps
• (optional) V4L2 1.0.1

Required hardware

• Webcam or some other video-source.
• (optional) GPS Module BU-353

阅读全文……