Machine Learning for Pedestrian Detection
How does a Smart Assistance System detects Pedestrian?
Phases Object Segmentation Feature Extraction Classification Get foreground image and segment Extract the relevant features in the image Classifies the images into respective classes
Feature Extraction Using HAAR Transform Each Rectangle Bar represents a Feature subtraction of sum of rectangle grey scale of black block and white block gives the intensity of the pixel
Classification with Adaboost and SVM
Support Vector machine margin Others SVM
Training Data Positive Samples Negative Samples
Analysis IS Pedestrian?Predicted : YesPredicted : No Total Positive Samples: P True Positive : TPFalse Negative Total Negative Samples: N False Positive : FP True Negative: TN Accuracy (AC): (TP+TN)/(P+N) Detection Rate (DR): TP/P False Alarm Rate : FP/N
Comparison of Results Classifier Data Sets(P=100,N=500) 1 2 Single SVM AR(%) DR(%) FPR(%) 0.00 Cascade- Adaboost-SVM AR(%) DR(%) FPR(%) 0.00
Classifier Comparison Data Set Cascade Classifier SVM Number of SV’s Comparison of number of support vectors between cascade classifier and SVM
Let’s watch EmlaEA767A5K-hduKQRx&index=14 Volvo – S60 Pedestrian Detection System
Other Applications Surveillance Systems Starts Recording after detecting the Pedestrians. Reduce the space to store the videos.
Contd.. Human Robot Interactions
Any Queries ??