Seminar Project | Volume Estimation

Transcript

1. PROJECT SEMINAR THEME Volume Estimation of Real world objects Joel Vilanilam Zachariah Roll No. 30 | CSU 161 30 | MDL16CS059 Govt. Model Engineering College
2. BASE PAPER MUSEFood OTHER APPROACHES Kabaq WHY THIS THEME? ARomapp KEY HIGHLIGHTS
3. PART 0: PROLOGUE OUR PURSUIT TO REDUCE FOOD WASTAGE
4. PART 1: DEPTH ESTIMATION TECHNIQUES DERIVING 3D FROM 2D
5. BASE PAPER MUSEFood MULTI-SENSOR-BASED FOOD VOLUME ESTIMATION ON SMARTPHONES
6. PART 2: INTRODUCTION THE APPROACH OF THIS METHOD
7. FOOD SEGMENTATION FROM BACKGROUND (1) CNN + GrabCut (2) Point Matching RELATIVE FOOD SIZE TO ACTUAL FOOD SIZE (1) Place reference object (2) Depth image from special device
8. PART 3: RELATED WORK
9. FOOD ITEM SEGMENTATION Location of Food item in image ACTUAL SIZE SCALING Reference Object OR Train to detect FOOD MODELING Deriving Deep maps from deep Neural Networks OR Stereo Matching
10. PART 4: METHODS
11. OVERVIEW 1) Sensing 2) Data Processing 3) Data Aggregation SENSING 1) Food Image sensing 2) Audio sensing DATA AGGREGATION 1) Actual Size Scaling 2) Food Modeling & Volume Calculation DATA PROCESSING 1) Audio Signal Processing for distance measurement 2) Food Item Segmentation
12. SENSING Take a top-down as well as a side view photograph Ensure they are respectively parallel and perpendicular to surface Echo ranging is used with a Maximum Length Sequence (MLS) MLS is a pseudorandom binary sequence (contains 0's and 1's) Length of MLS = 2^n -1 Start recording the effect of emiting audio signal while taking photographs. Stop recording when done with capturing photographs. 1) FOOD IMAGE SENSING 2) AUDIO SENSING
13. DATA PROCESSING 1) AUDIO SIGNAL PROCESSING FOR DISTANCE MEASUREMENT
14. DATA PROCESSING Two Tasks: Food segmentation Task Container Classification Task 2) FOOD ITEM SEGMENTATION Hence, this is a Multi-task deep learning problem.
16. DATA AGGREGATION 1) ACTUAL SIZE SCALING 2) FOOD MODELING AND VOLUME CALCULATION
17. PART 5: EVALUATION
18. MLS RANGING 1) Ranging Accuracy 2) Ranging Robustness FOOD IMAGE SEGMENTATION 1) Dataset 2) Experimental Set Up 3) Evaluation Metric 4) Performance FOOD VOLUME ESTIMATION Testing it out
21. PART 6: CONCLUSION
23. PART 7: EPILOGUE REAL WORLD IMPLEMENTATIONS
25. THANK YOU