Frédéric Chausse

A model-driven approach for real-time road recognition

Abstract. This article describes a method designed to detect and track road edges starting from images provided by an on-board monocular monochromic camera. Its implementation on specific hardware is also presented in the framework of the VELAC project. The method is based on four modules: (1) detection of the road edges in the image by a model-driven algorithm, which uses a statistical model of the lane sides which manages the occlusions or imperfections of the road marking – this model is initialized by an off-line training step; (2) localization of the vehicle in the lane in which it is travelling; (3) tracking to define a new search space of road edges for the next image; and (4) management of the lane numbers to determine the lane in which the vehicle is travelling. The algorithm is implemented in order to validate the method in a real-time context. Results obtained on marked and unmarked road images show the robustness and precision of the method.

Accurate road following and reconstruction by computer vision

This paper presents a method designed to track and to recover the three-dimensional (3-D) shape of a road by computer vision. The method is based first upon an accurate detection algorithm which provides a reliable estimation of the roadside in the image. This algorithm works by recursive updating of a statistical model of the lane obtained by an off-line training phase. Once the sides have been located, a reconstruction algorithm computes the vehicle location on its lane, the 3-D shape of the road, and gives both the sides location and their confidence interval for the next image. The detection algorithm then looks for the roadside in this interval in order to limit the computational times, which are about 30-150 ms on a HP workstation.

Vehicle localization on a digital map using particles filtering

Localization is an important functionality for the navigation of intelligent vehicles. It is usually done using several kinds of sensors (proprioceptive, GPS, camera). All the data are uncertain and even momentarily unavailable (GPS in urban areas for example). A data fusion process is necessary for sensors data to compensate one each other. We propose here to combine GPS absolute localization with data computed by a vision system giving the position and orientation of the vehicle on the road. This last local information is transformed into a global reference using a map of the environment. The localization parameters are estimated using a particles filter making it possible to manage multimodal estimations (the vehicle can be on the left lane or on the right one for example). Many results have been obtained in real time and on real roads by implementing this solution in an experimental vehicle. The best standard deviation reached is 48 cm along the road axis and 8 cm along the axis normal to the road.

Localization of a Group of Communicating Vehicles by State Exchange

This paper considers the problem of cooperative localization of an heterogeneous group of road vehicles. Each vehicle can be equipped with proprioceptive and exteroceptive sensors enabling it to localize itself in its environment and also to localize (but not to identify) the other members of the group. Localization information can be exchanged between the vehicles through a wireless communication device. Every member of the group maintains (if possible) an estimation of the state of its environment and transmits it (if possible) to its neighbors. The global state of the environment is obtained by fusing the environment states of the vehicles. This fusion is based on extended Kalman filter where the poses of the detected vehicles are represented by a single system. The proposed approach takes into account the sensor constraints such as data unavailability and delays

A fast and robust vision based road following algorithm

This article presents a fast and robust method designed to detect and track a road lane from images provided by an on-board monocular monochromatic camera. The detection method is based upon a model driven algorithms. It uses a statistical model of the lane which permits to manage the occlusions or imperfections of road marking. This model is obtained by an off-line training phase. The detections are made in optimal interest zones deduced from the model. The tracking process permits to locate the vehicle on its lane and gives the confidence interval of the roadside for the next image. The method has been applied both on marked and unmarked roads images. The results obtained on image sequences of real road scenes show the robustness and precision of the proposed approach.

A real-time, multi-sensor architecture for fusion of delayed observations: application to vehicle localization

This paper presents a software framework called AROCCAM that was developed to design and implement data fusion applications. This architecture permits to build applications in a very short time unburdening the user of sensor communication. Moreover, it manages unsynchronized sensors and delayed observations in an elegant manner that permits the user to fuse those information easily, taking into account the environment perception date. In this paper, a fusion methodology for delayed observations is first presented in order to point the problem of latency periods in a fusion system. These latency periods are then taken into account within our embedded architecture needing only a little effort from user. Finally, benefits of AROCCAM architecture are demonstrated via a real-time vehicle localization experiment carried out with an outdoor robot

Real-time vehicle trajectory supervision on the highway

This article presents a method for real-time control of vehicle trajectory on a highway based on an on-board vision system using a single camera. The system has been designed to avoid damage (e.g., due to a sleepy driver). The method is based on the real-time extraction of the lateral vehicle location, which is used to determine the vehicle trajectory. Furthermore, the system is able to correct possible bad road tracking in order to have a complete autonomy. The whole system has been im plemented on a single TMS320C50 DSP-based card. Reliable results have been obtained on the highway over several hun dreds of kilometers at low and high speeds reaching more than 130 km/hr.

Cooperative Multi-Vehicle Localization

This paper considers the problem of cooperative localization of an heterogeneous group of road vehicles. Each vehicle is equipped with proprioceptive and exteroceptive sensors enabling it to localize itself in its environment and also to identify and localize the other members of the group. Localization information can be exchanged between the vehicles through a wireless communication device. Every member of the group maintains an estimation of the state of its environment and transmits it to its neighbors. The global state of the environment is obtained by fusing the environment states of the vehicles. This fusion is based on an extended Kalman filter where the group is represented by a single system which describes the state of every member. The proposed approach take into account sensor constraints as data unavailability and delays

Recovering the 3D shape of a road by on-board monocular vision

Deals with a method designed to recover the 3D geometry of a road from an image sequence provided by an on-board monocular monochromatic camera. It only requires the road edges to be detected in the image. The reconstruction process is able to compute (1) the 3D coordinates of the road axis points, (2) the vehicles position on its lane and (3) the prediction of the road edge localization in the next images of the sequence which is very helpful for the detection phase. It also computes the confidence intervals associated with the 3D parameters. The description of the method is followed by the presentation of its most significant results.

Real-time tracking with classifiers

Two basic facts motivate this paper: (1) particle filter based trackers have become increasingly powerful in recent years, and (2) object detectors using statistical learning algorithms often work at a near real-time rate. We present the use of classifiers as likelihood observation function of a particle filter. The original resulting method is able to simultaneously recognize and track an object using only a statistical model learnt from a generic database. Our main contribution is the definition of a likelihood function which is produced directly from the outputs of a classifier. This function is an estimation of calibrated probabilities P(class|data). Parameters of the function are estimated to minimize the negative log likelihood of the training data, which is a cross-entropy error function. Since a generic statistical model is used, the tracking does not need any image based model learnt inline. Moreover, the tracking is robust to appearance variation because the statistical learning is trained with many poses, illumination conditions and instances of the object. We have implemented the method for two recent popular classifiers: (1) Support Vector Machines and (2) Adaboost. An experimental evaluation shows that the approach can be used for popular applications like pedestrian or vehicle detection and tracking. Finally, we demonstrate that an efficient implementation provides a real-time system on which only a fraction of CPU time is required to track at frame rate.