kankan-sarkar/pedestriandetector

PedestrianDetector Implementation using SVM and Random Forest. SVM , RDF , PEDESTRIAN DETECTION

PedestrianDetector Implementation using SVM and Random Forest

File Execution Sequence

Hog Implementation(Histogram of oriented Gradient)

Gradient Calculation(Gx, Gy, Magnitude , and orientation)

HOG output Visualization(my implementation vs Skimage HOG implementation). Greyscale pixel values are used to represent magnitudes and arrows to represent orientations

Generate positive and negative instance. The positive and negative instances which are generated are kept in two separate directories

• …\data\train_pos
• …\data\train_neg

Class-balanced random train/test split using sklearn.model_selection methods

Train an Sklearn random decision forest on feature vectors generated using your HOG function from (1)

Grid-based cross validation

0.9969 (+/-0.0083) for {'max_depth': 5, 'n_estimators': 200}
0.9979 (+/-0.0051) for {'max_depth': 5, 'n_estimators': 400}
0.9969 (+/-0.0083) for {'max_depth': 5, 'n_estimators': 600}
0.9969 (+/-0.0083) for {'max_depth': 5, 'n_estimators': 800}
0.9969 (+/-0.0083) for {'max_depth': 5, 'n_estimators': 1000}
0.9990 (+/-0.0041) for {'max_depth': 10, 'n_estimators': 200}
0.9990 (+/-0.0041) for {'max_depth': 10, 'n_estimators': 400}
0.9990 (+/-0.0041) for {'max_depth': 10, 'n_estimators': 600}
0.9990 (+/-0.0041) for {'max_depth': 10, 'n_estimators': 800}
0.9979 (+/-0.0051) for {'max_depth': 10, 'n_estimators': 1000}

• max_depth -> Change in max depth from 5 to 10 increased the grip score

Training an Sklearn support vector classifier on feature vectors generated using your HOG function from (1)

Use grid-based cross validation to optimize the support vector classifier hyper parameters

0.9884 (+/-0.0129) for {'C': 1, 'gamma': 0.01, 'kernel': 'rbf'}
0.9705 (+/-0.0124) for {'C': 1, 'gamma': 0.001, 'kernel': 'rbf'}
0.9457 (+/-0.0156) for {'C': 1, 'gamma': 0.0001, 'kernel': 'rbf'}
0.9932 (+/-0.0099) for {'C': 10, 'gamma': 0.01, 'kernel': 'rbf'}
0.9903 (+/-0.0148) for {'C': 10, 'gamma': 0.001, 'kernel': 'rbf'}
0.9705 (+/-0.0124) for {'C': 10, 'gamma': 0.0001, 'kernel': 'rbf'}
0.9932 (+/-0.0099) for {'C': 15, 'gamma': 0.01, 'kernel': 'rbf'}
0.9932 (+/-0.0099) for {'C': 15, 'gamma': 0.001, 'kernel': 'rbf'}
0.9742 (+/-0.0127) for {'C': 15, 'gamma': 0.0001, 'kernel': 'rbf'}
0.9942 (+/-0.0113) for {'C': 1, 'kernel': 'linear'}
0.9942 (+/-0.0113) for {'C': 10, 'kernel': 'linear'}
0.9942 (+/-0.0113) for {'C': 15, 'kernel': 'linear'}

• It is both dependent on C and gamma parameter

Results