Super Effective SVM-based Techniques Towards Autonomous Vehicles For Improving Collision Avoidance

This thesis focuses on advancing the vehicle detection,distance estimation,lane detection and speed estimation in autonomous driving.An autonomous vehicle requires a lot of walking Unexpected and dynamic environment.Therefore,a strong cognitive system is necessary.This work proposes novel techniques,support vector machine-based techniques capable of finding lane on the road and generating accurate bounding boxes around detected vehicle and estimate distance.Camera calibration,which would help us undistort the images for obtaining better accuracy.Projective transform,which will be used to obtain top-view of the road.Estimating resolution,which would help us transform pixels into meters or feet.To do that,the standardized minimal width of lane of 12 feet(or 3.6576 meters)will be used.Once the lane found there is additional requirement to calculate the curvature of the lane and position of the car relative to the lane.The second stage is to create a classifier which classifies car against non-cars.To do so the dataset is needed,the dataset is a combination of KITTI vision benchmark suite and GTI vehicle image database.To build a classifier,first,the features have to be identified.The features that are going to be used is a mixture of histograms,full images,and HOG-s.Second Build and train the classifier,the classifier used is linear support vector classifier.After that,the dataset is split into a training and test set,where the test set is 10% of all the data.The final accuracy obtained on the test set was 99.55%.To measure the distance the midpoint of the lower edge of a bounding box is used.The bounding box is averaged over last 21 frames.The goal of the project was to find vehicle,estimate distance and highlight the road lane on the video recorded from the car.This project can be useful in reducing the risk of accidents due to human error,many more benefits.