Car Detection

Overview

This project focus on the detection and recognition of cars in different perspective views and has the following associated paper:

Multiview object recognition using Bag of Words approach

Abstract: Multiview object detection and classification plays a critical role in robust image recognition systems, and can be applied in a multitude of applications, ranging from simple monitoring to advanced tracking. In this paper it is analyzed the usage of the Bag of Words model to efficiently detect and recognize objects that can appear in different scales, orientations and even from different perspective views. This approach relies in image analysis techniques, such as feature detection, description and clustering, in order to be able to recognize the target object even if it is present in cluttered environments. For supporting the recognition in different perspective views, machine learning techniques are used to build a model of the target objects. This model can then be employed to successfully recognize if an instance of the target object is present in an image. For pinpointing the location of the target object, a sliding window method is used in conjunction with dynamic thresholding. The recognition system was tested with several configurations of feature detectors, descriptors and classifiers, and achieved an accuracy of 87% when recognizing cars from 177 test images.

Results

Fig. 1 - Effect of preprocessing (right) in the original image (left)

Fig. 2 - Target objects ground truth masks

Fig. 3 - Results obtained with STAR detector, SIFT extractor, FLANN matcher and ANN classifier

Fig. 4 - Results with partially occluded objects obtained with STAR detector, SURF extractor, FLANN matcher and SVM classifier

Fig. 5 - Results obtained with STAR detector, FREAK extractor, FLANN matcher and SVM classifier

Fig. 6 - Results obtained with STAR detector, SIFT extractor, FLANN matcher and SVM classifier

Fig. 7 - Results obtained with SURF detector, SURF extractor, FLANN matcher and ANN classifier

Fig. 8 - Results obtained with FAST detector, SURF extractor, FLANN matcher and ANN classifier

Fig. 9 - Results obtained with ORB detector, ORB extractor, FLANN matcher and ANN classifier

Releases

Windows 8.1 release

Building and developing

The setup instructions on how to build and develop in Visual Studio is available here