Predicting distant future trajectories of agents in a dynamic scene is challenging because the future trajectory of an agent is affected not only by their past trajectory but also the scene contexts. To tackle this problem, the authors propose a model based on recurrent neural networks, and a novel method for training this model. The proposed model is based on an encoder–decoder architecture where the encoder encodes inputs (past trajectory and scene context information), while the decoder produces a future trajectory from the context vector given by the encoder. To make the proposed model better utilise the scene context information, the authors let the encoder predict the positions in the past trajectory and a reward function evaluate the positions along with the scene context information generated by the positions. The reward function, which is simultaneously trained with the proposed model, plays the role of a regulariser for the model during the simultaneous training. The authors evaluate the proposed model on several public benchmark datasets. The experimental results show that the prediction performance of the proposed model is greatly improved by the proposed regularisation method, which outperforms the‐state‐of‐the‐art models in terms of accuracy.

This paper considers a method for detection of road surface markings using a camera mounted on top of a vehicle. The detection is done with an orientation-aware detector based on a convolutional neural network. To successfully detect the orientation and position of road surface markings, the input frontal image is converted to a bird’s-eye view image using inverse perspective matching. Synthetic image dataset is constructed with aid of MSER (maximally stable extremal regions) algorithm to solve data imbalance problem. The detector is trained to estimate orientations of the detected objects in addition to the class labels and positions. Pretrained DenseNet based YOLOv2 model is modified to detect rotated rectangles with an additional cost function and new efficient IOU (intersection of union) measure. Instead of directly estimating the orientation angle of the road surface markings, probabilistic estimation is done with quantized angular bins. Benchmark dataset is formulated for evaluation and the experimental results showed that the considered algorithm provides promising result while running in a real-time.

In this paper, we propose a future trajectory prediction framework based on recurrent neural network (RNN) and maximum margin inverse reinforcement learning (IRL) for the task of predicting future trajectories of agents in dynamic scenes. Given the current position of a target agent and the corresponding static scene information, a RNN is trained to produce the next position, which is the closest to the true next position while maximizing the proposed reward function. The reward function is also trained at the same time to maximize the margin between the rewards from the true next position and its estimate. The reward function plays the role of a regularizer when training the parameters of the proposed network so the trained network is able to reason the next position of the agent much better. We evaluated our model on a public KITTI dataset. Experimental results show that the proposed method significantly improves the prediction accuracy compared to other baseline methods.

In this paper, we propose an accurate lane-level map building method using low-cost sensors such as cameras, GPS and in-vehicle sensors. First, we estimate the ego-motion from the stereo camera and the in-vehicle sensors, and globally optimize the accurate vehicle positions by fusion with the GPS data. Next, we perform lane detection on every image frame in the camera. Lastly, we repeatedly accumulate and cluster the detected lanes based on the accurate vehicle positions, and perform polyline fitting algorithm. The polyline fitting algorithm follows a variant of the Random Sample Consensus (RANSAC) algorithm, which particularly solves the multi-line fitting problem. This algorithm can expand the lane area and improve the accuracy at the same time by repeatedly driving the same road. We evaluated the lane-level map building on two types of roads: a proving ground and a real driving environment. The lane map accuracy verified at the proving ground was 9.9982cm of the CEP error.

We present a vehicle re-identification method for a parking lot toll system. Given a probe image captured from one camera installed in the entrance of a parking lot, re-identification is the method of identifying a matching image from a gallery set constructed from different cameras in the exit region. This method is especially useful when the license plate recognition fails. Our method is based on a convolutional neural network (CNN) which is a variant of multilayer perceptron (MLP). An input image of the CNN model is cropped by the license plate detection (LPD) algorithm to eliminate the background of an original image. To train a vehicle re-identification model, we adopt the pre-trained models which showed the outstanding results in the ImageNet [1] challenge from 2014 to 2015. Then, we fine-tune one of the models (GoogLeNet [2]) for a car’s model recognition task using a large-scale car dataset [3]. This fine-tuned model is utilized as a feature extractor. Cosine function is used to measure the similarity between a probe and a gallery. To evaluate the performance of our method, we create two datasets: ETRI-VEHICLE-2016-1 and ETRI-VEHICLE2016-2. The experimental result reveals that the proposed technique can achieve promising results.

This paper discusses an efficient method for generating height-grid maps, which are widely utilized in developing mobile robots and autonomous vehicles. The proposed method consists of four steps. First, inverse perspective images (IPI) are obtained from wide-angle cameras equipped with a fish-eye lens. Second, dense motion stereo is performed using IPI. Next, depth information is calculated and mapped onto the grid map. Lastly, a fusion of the grid maps is executed to enhance the quality of the map using a modified temporal median filter. The method can efficiently collect height and position information about objects relative to the road surface. The reason is that wide-angle images are converted to IPI to remove the perspective distortion of the road surface, and the depth is calculated using dense motion stereo from IPI. Moreover, the efficiency of the proposed fusion method is significantly higher than that of previous methods. The proposed method is very simple, but the quality that could be obtained is similar to the quality produced by the conventional technique, and the building speed showed dramatically fast performance of up to 200 fps in a single core CPU. In addition, we used a variety of test data and showed that the result verifies the validity of the proposed algorithm.

This note describes a synchronization method for use of the multichannel display and the associated clustered PC rendering systems, within multi user Virtual Reality online games with real time frame rates. It provides a real time surround game display to render the very complex 3D graphics objects. Our system is composed of Multi-channel Distributor, and Visualization Clients. Visual client system can be connected to multi monitors or projectors. With Multi-channel distributor, an application can be configured, at runtime, to span any number of visuals to makeup a complete surround and to manage the user game event. All visualization clients produced are frame synchronized to ensure visual uniformity. However, tightly synchronized PC cluster may cause a draw back in its rendering speed. So, we propose a dual scene graph management method to adaptively synchronize. One scene graph is conventionally organizes and control the rendering of its constituent objects. And the other is spatially grouping on the geometric objects found at the edge of two adjacent channels between of the clustered PCs and or not. In this paper describes the method of a construction dual-scene graph based clustered rendering system for scalable display in order to generate high-resolution images at real time frame rates. And also we are planning to develop a multi-projector display system which calibrates itself automatically regardless its shape of display surface and which provides a user with a seamless game display. Our results are well suited for surround 3D game displays such as multi-projector or multi-monitor running on a PC cluster.