Low Resolution Lidar-Based Multi-Object Tracking for Driving Applications

Abstract

Abstract. Vehicle detection and tracking in real scenarios are key components to develop assisted and autonomous driving systems. Lidar sensors are specially suitable for this task, as they bring robustness to harsh weather conditions while providing accurate spatial information. However, the resolution provided by point cloud data is very scarce in comparison to camera images. In this work we explore the possibilities of Deep Learning (DL) methodologies applied to low resolution 3D lidar sensors such as the Velodyne VLP-16 (PUCK), in the context of vehicle detection and tracking. For this purpose we developed a lidar-based system that uses a Convolutional Neural Network (CNN), to perform pointwise vehicle detection using PUCK data, and Multi-Hypothesis Extended Kalman Filters (MH-EKF), to estimate the actual position and velocities of the detected vehicles. Comparative studies between the proposed lower resolution (VLP-16) tracking system and a high-end system, using Velodyne HDL-64, were carried out on the Kitti Tracking Benchmark dataset. Moreover, to analyze the influence of the CNN-based vehicle detection approach, comparisons were also performed with respect to the geometric-only detector. The results demonstrate that the proposed low resolution Deep Learning architecture is able to successfully accomplish the vehicle detection task, outperforming the geometric baseline approach. Moreover, it has been observed that our system achieves a similar tracking performance to the high-end HDL-64 sensor at close range. On the other hand, at long range, detection is limited to half the distance of the higher-end sensor.

Object classification integrating results of each scan line with low‐resolution LIDAR

“…We used LIDAR having eight scan lines, and tried to classify pedestrians, bicycles, motorbikes, cars, and other objects. In addition, we verified that the proposed method had advantages over scan line missing by comparing the method of converting LIDAR data to depth map and recognition with 2D CNN [22]. In ref.…”

“…To verify robustness against missing scan lines, we compared the proposed method with 2DCNN [22] and fully connected NN. In this 2DCNN method, the measured LIDAR data were converted to 8 × 3500 depth map.…”

“…In ref. [22], del Pino et al used 2DCNN for car detection. We applied 2DCNN to our multiclass classification task as a comparison method.…”

Review on Common Techniques for Urban Environment Video Analytics

“…Contrary to the traditional techniques explained above, deep learning-based techniques have been used in the last decade to significantly increase object tracking performance. Del Pino et al [del Pino et al 2017] propose the use of convolutional neural networks to estimate the real position and speeds of detected vehicles. On the other hand, Liu et al [Liu et al 2016] propose a deep learning-based progressive vehicle reidentification approach, which uses a convolutional neural network to extract appearance attributes as the first filter, and a Siamese neural network-based license plate verification as the second filter.…”

Road Segmentation on Low Resolution Lidar Point Clouds for Autonomous Vehicles

“…These differences directly affect the performance of deep learning models that learn representations for different tasks, such as semantic segmentation and object detection. Low-resolution 16 layer LIDARs have been recently compared with 64 layer LIDARs (del Pino et al, 2017) to evaluate the degradation in detection accuracy especially w.r.t distance.…”