Automatic road environment classification 20121002

452
-1

Published on

2012.10.04

Published in: Technology
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total Views
452
On Slideshare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Automatic road environment classification 20121002

  1. 1. Advisor : Yin-Fu Huang Automatic Road Environment Classification Intelligent Transportation Systems, IEEE Transactions on Student : Chen-Ju Lai 1
  2. 2. Outline Introduction Road image subregions Feature extraction and representation Feature classification Experimental result Conclusion 2
  3. 3. Introduction Input video frame :640X480 resolution Standard digital video camera(30-fps video) Feature extraction :color ,texture Classification method :k-NN and ANN Four class problem , accuracy : ~80% {off−road, urban, major/trunk road,multilane motorway/carriageway} Two class problem , accuracy : ~90% {off−road, on−road} A near real time classification rate of 1Hz. (i.e.,one frame classification per second) 3
  4. 4. Introduction Colour based off-road environment and terrain type classification[3] Texture and neural network for road segmentation[7] Detection and classification of highway lanes using vehicle motion trajectories[31] 4
  5. 5. Road image subregions5
  6. 6. Feature extraction and representation Color Features  Choice color space 6
  7. 7. Feature extraction and representation Color Features  Each such channel of each subregion as the normalized histogram distribution.  Mean , standard deviation , and entropy. A given color value (indexed k = 1, . . . , L) occurs with probability pk  Each color channel is summarized as a color feature vector of combining the histogram (quantized to 10 “bins”)  13-value , 7 channel , 91-D color descriptor for each image frame. 7
  8. 8. Feature extraction and representation Texture Features  grey-level co-occurrence matrix (GLCM) statistics  localized orientation is defined with as {N,S, E,W,NW,NE,SW,SE} Original image Co-occurrence Matrix 8
  9. 9. Feature extraction and representation Texture Features  grey-level co-occurrence matrix (GLCM) statistics Co-occurrence Matrix Stochastic Matrix 9
  10. 10. Feature extraction and representation Texture Features  grey-level co-occurrence matrix (GLCM) statistics 10
  11. 11. Feature extraction and representation Texture Features  grey-level co-occurrence matrix (GLCM) statistics M(i, j) is the (i, j)th entry in GLCM M with dimension(colsx rows) 11
  12. 12. Feature extraction and representation Texture Features  grey-level co-occurrence matrix (GLCM) statistics horizontal standard deviation and mean (σI, μI ), vertical standard deviation and mean (σJ, μJ ). 12
  13. 13. Feature extraction and representation Texture Features  Gabor filters allow the study of the localized spatial distribution of the texture.  The magnitude of the Gabor filter response identifies varying local texture frequencies and orientations in the image.  Use for the extraction of more gradual (low-frequency) textures and more generally create discriminative texture descriptors. 13
  14. 14. Feature extraction and representation Texture Features  The use of only the (N,E) GLCM directions is within this visual discriminatory context.  The Gabor filter in use is itself summarized as a quantized histogram (10 “bins”)  Mean, standard deviation, and entropy  23-value ,3 subregions , 69-D texture feature descriptor for each image frame. 14
  15. 15. Feature extraction and representation Edge-Derived Features  Three additional edge-derived features specific to the road- edge subregion.  Use Canny edge detector  The set of edges, connected contours , and straight line detected is then summarized by the entropy. 15
  16. 16. Feature Classification A 163-D combined feature vector per image frame. K-NN and ANN Training set : 800 image frames (200 per class ,for four classes) Four-class  classes = {off−road, urban, major/trunk road, multilane motorway/carriageway} Two-class  classes = {off−road, on−road}  {on−road} = {{urban} ∪ {major/trunk road} ∪ {multilane motorway/carriageway}} 16
  17. 17. Experimental result Two test video sequences Video sequence Duration Consist of Video sequence 1 40-s 10-s segments of {off-road, urban, major/trunk road, multilane motorway/carriageway} Video sequence 2 50-s 10-s segments of {urban, major/trunk road, multilane motorway/carriageway} Test image frame 20-s segments of {off-road} Road environment 600(20-s) urban 600(20-s) major/trunk road 30-fps video 600(20-s) multilane motorway/carriageway 900(30-s) off-road 17
  18. 18. Experimental result K-NN classificationFig. 2. Video sequence 1. Classification results using k-NN varyingparameter k. 18
  19. 19. Experimental result K-NN classificationFig. 3. Video sequence 2. Classification results using k-NN varyingparameter k. 19
  20. 20. Experimental result ANN classification  We employ a classical two-layer network topology with H hidden nodes.  163 input node , 2 or 4 output node (two class or four class).  The ANN is trained using I iterations.  The general range of parameter H ={10, . . . , 60} and I = {150, . . . , 700}. 20
  21. 21. Experimental result ANN classification 21
  22. 22. Experimental result ANN classification 22 Fig. 4. Examples of successful ANN classification for four road environments
  23. 23. Experimental result ANN classification 23 Fig. 5. Examples of successful ANN classification for two road environments (ANN configuration: H = 15 Nodes; I = 200).
  24. 24. Experimental result Extended Sequence Results  full video sequences (representing the complete set of viable data gathered over several hours in varying environments) 24
  25. 25. Experimental result Misclassification 25
  26. 26. Conclusion An ANN classifier gives ∼90%–97% successful classification for two class ; ∼80%–85% for four-class. A k-NN classifier implies the inherent feature overlap within the current feature space ,so it’s difficult to classify. Future work  Subregion optimization  Alternative computationally efficient texture measures  The efficient of varying weather and lighting condition on performance 26

×