Bogdan Stanciulescu

Person re-identification in multi-camera system by signature based on interest point descriptors collected on short video sequences

We present and evaluate a person re-identification scheme for multi-camera surveillance system. Our approach uses matching of signatures based on interest-points descriptors collected on short video sequences. One of the originalities of our method is to accumulate interest points on several sufficiently time-spaced images during person tracking within each camera, in order to capture appearance variability. A first experimental evaluation conducted on a publicly available set of low-resolution videos in a commercial mall shows very promising inter-camera person re-identification performances (a precision of 82% for a recall of 78%). It should also be noted that our matching method is very fast: ~ 1/8s for re-identification of one target person among 10 previously seen persons, and a logarithmic dependence with the number of stored person models, making re- identification among hundreds of persons computationally feasible in less than ~ 1/5 second.

Real-time traffic sign recognition in three stages

Traffic Sign Recognition (TSR) is an important component of Advanced Driver Assistance Systems (ADAS). The traffic signs enhance traffic safety by informing the driver of speed limits or possible dangers such as icy roads, imminent road works or pedestrian crossings. We present a three-stage real-time Traffic Sign Recognition system in this paper, consisting of a segmentation, a detection and a classification phase. We combine the color enhancement with an adaptive threshold to extract red regions in the image. The detection is performed using an efficient linear Support Vector Machine (SVM) with Histogram of Oriented Gradients (HOG) features. The tree classifiers, K-d tree and Random Forest, identify the content of the traffic signs found. A spatial weighting approach is proposed to improve the performance of the K-d tree. The Random Forest and Fishers Criterion are used to reduce the feature space and accelerate the classification. We show that only a subset of about one third of the features is sufficient to attain a high classification accuracy on the German Traffic Sign Recognition Benchmark (GTSRB).

Real-Time Traffic-Sign Recognition Using Tree Classifiers

Traffic-sign recognition (TSR) is an essential component of a driver assistance system (DAS), providing drivers with safety and precaution information. In this paper, we evaluate the performance of k-d trees, random forests, and support vector machines (SVMs) for traffic-sign classification using different-sized histogram-of-oriented-gradient (HOG) descriptors and distance transforms (DTs). We also use the Fishers criterion and random forests for the feature selection to reduce the memory requirements and enhance the performance. We use the German Traffic Sign Recognition Benchmark (GTSRB) data set containing 43 classes and more than 50 000 images.

Traffic sign classification using K-d trees and Random Forests

In this paper, we evaluate the performance of K-d trees and Random Forests for traffic sign classification using different size Histogram of Oriented Gradients (HOG) descriptors and Distance Transforms. We use the German Traffic Sign Benchmark data set [1] containing 43 classes and more than 50,000 images. The K-d tree is fast to build and search in. We combine the tree classifiers with the HOG descriptors as well as the Distance Transforms and achieve classification rates of up to 97% and 81.8% respectively.

Segmentation masks for real-time traffic sign recognition using weighted HOG-based trees

Traffic sign recognition is one of the main components of a Driver Assistance System (DAS). This paper presents a real-time traffic detection and classification approach of both circular and triangular signs. The system consists of three stages: 1) an image segmentation to reduce the search space, 2) a HOG-based Support Vector Machine (SVM) detection to extract both round and triangular traffic signs, and 3) a tree classifier (K-d tree or Random Forests) to identify the signs found. The methodology is tested on images under bad weather conditions and poor illumination. The image segmentation based on the enhancement of the red color channel improves the detection precision significantly achieving high recall rates and only a few false alarms. The tree classifiers also achieve high classification rates.

COMBINING ADABOOST WITH A HILL-CLIMBING EVOLUTIONARY FEATURE SEARCH FOR EFFICIENT TRAINING OF PERFORMANT VISUAL OBJECT DETECTORS

This paper presents an efficient method for automatic training of performant visual object detectors, and its successful application to training of a back-view car detec- tor. Our method for training detectors is adaBoost applied to a very general family of visual features (called “control-point” features), with a specific feature-selection weak-learner: evo-HC, which is a hybrid of Hill-Climbing and evolutionary-search. Very good results are obtained for the car-detection application: 95% positive car detection rate with less than one false positive per image frame, computed on an independant validation video. It is also shown that our original hybrid evo-HC weak-learner allows to obtain detection performances that are unreachable in rea- sonable training time with a crude random search. Finally our method seems to be potentially efficient for training detectors of very different kinds of objects, as it was already previously shown to provide state-of-art performance for pedestrian-detection tasks.

Rail extraction technique using gradient information and a priori shape model

This paper presents a comparative study of different rail detection techniques as well as a new method based on an efficient algorithm without any empirical thresholds. The main problem with rail detection is that both the track-bed and the exterior conditions (weather/light conditions) vary along the path. On the other hand, there are properties that can be exploited to improve the conventional lane detection. We present an edge detection based on the estimated position of the rails that follows the rail edges upwards in the image, determining a free-from-obstacles space. The existing techniques are also analyzed and compared.

Real-time traffic sign recognition using spatially weighted HOG trees

Traffic sign recognition is one of the main components of a Driver Assistance System (DAS). This paper presents a real-time traffic sign recognition system. It consists of three stages: 1) an image segmentation using red color enhancement to reduce the search space, 2) a HOG-based Support Vector Machine (SVM) detection to extract the traffic signs, and 3) a tree classifier (K-d tree or Random Forests) to identify the signs found. This methodology is tested on images under bad weather conditions and poor illumination. The tree classifiers achieve high classification rates for the German Traffic Sign Recognition Benchmark and the ETH 80 dataset. The K-d tree classification is improved by introducing a Gaussian spatial weighting to favor the interior blocks of the HOG descriptors.

Rail and turnout detection using gradient information and template matching

This paper presents a railway track and turnout detection algorithm which is not based on an empirical threshold. The railway track extraction is based on an edge detection using the width of the rolling pads. This edge detection scheme is then used as an input to the RANSAC algorithm to determine the model of the rails. The turnout detection scheme is based on the Histogram of Oriented Gradient (HOG) and Template Matching (TM). The results show (i) reliable performance for our railway track extraction scheme and (ii) a correction rate of 97.31 percent for the turnout detection scheme using a Support Vector Machine (SVM) classifier.

Turnout detection and classification using a modified HOG and template matching

This paper presents a railway track and turnout detection and turnout classification algorithm. The railway track extraction is based on an edge detection using the width of the rolling pads. This edge detection scheme is then used as an input to the RANSAC algorithm to determine the model of the rails knowing their gauge. The turnout detection scheme is based on the Histogram of Oriented Gradient (HOG) and Template Matching (TM). The turnout classification is based on HOG. The detection results show (i) reliable performance for our railway track extraction scheme; (ii) a correction rate of 97.31 percent for the turnout detection scheme using a Support Vector Machine (SVM) classifier. The turnout classification has correction rate of 98.72 percent using SVM.