Search

Conventional deep learning based methods for object detection require a large amount of bounding box annotations for training, which is expensive to obtain such high quality annotated data. Few-shot object detection, which learns to adapt to novel classes with only a few annotated examples, is very challenging since the fine-grained feature of novel object can be easily overlooked with only a few data available. In this work, aiming to fully exploit features of annotated novel object and capture fine-grained features of query object, we propose Dense Relation Distillation with Context-aware Aggregation (DCNet) to tackle the few-shot detection problem. Built on the meta-learning based framework, Dense Relation Distillation module targets at fully exploiting support features, where support features and query feature are densely matched, covering all spatial locations in a feed-forward fashion. The abundant usage of the guidance information endows model the capability to handle common challenges such as appearance changes and occlusions. Moreover, to better capture scale-aware features, Context-aware Aggregation module adaptively harnesses features from different scales for a more comprehensive feature representation. Extensive experiments illustrate that our proposed approach achieves state-of-the-art results on PASCAL VOC and MS COCO datasets. Code will be made available at https://github.com/hzhupku/DCNet., Accepted by CVPR2021

Effective video-based detection methods are of great importance to intelligent transportation systems (ITS), and here we propose a method to localize and label objects. The method is able to detect pedestrians and bicycle riders in a complex scene. Our method is inspired by the common fate principle, which is a mechanism of visual perception in human beings, and which states tokens moving or functioning in a similar manner tend to be perceived as one unit. Our method embeds the principle in an Implicit Shape Model (ISM). In our method, keypoint-based object parts are firstly detected and then grouped by their motion patterns. Based on the grouping results, when the object parts vote for object centers and labels, each vote belonging to the same object part is assigned a weight according to its consistency with the votes of other object parts in the same motion group. Afterwards, the peaks, which correspond to detection hypotheses on the Hough image formed by summing up all weighted votes, become easier to find. Thus our method performs better in both position and label estimations. Experiments show the effectiveness of our method in terms of detection accuracy.

In our previous work, we reported a system that monitors an intersection using a network of horizontal laser scanners. This paper focuses on an algorithm for moving-object detection and tracking, given a sequence of distributed laser scan data of an intersection. The goal is to detect each moving object that enters the intersection; estimate state parameters such as size; and track its location, speed, and direction while it passes through the intersection. This work is unique, to the best of the authors' knowledge, in that the data is novel, which provides new possibilities but with great challenges; the algorithm is the first proposal that uses such data in detecting and tracking all moving objects that pass through a large crowded intersection with focus on achieving robustness to partial observations, some of which result from occlusions, and on performing correct data associations in crowded situations. Promising results are demonstrated using experimental data from real intersections, whereby, for 1063 objects moving through an intersection over 20 min, 988 are perfectly tracked from entrance to exit with an excellent tracking ratio of 92.9%. System advantages, limitations, and future work are discussed.

In this research, a novel system for monitoring an intersection using a network of single-row laser range scanners (subsequently abbreviated as "laser scanner") and video cameras is proposed. Laser scanners are set on the road side to profile an intersection horizontally from different viewpoints. The contour points of moving objects are captured at a certain horizontal plane with a high scanning rate (e.g., 37 Hz). A laser-based processing algorithm is developed, thus the moving objects entered the intersection are detected and tracked to estimate their state parameters, such as: location, speed, and direction at each time instance. In addition, laser data and processing results are forwarded to an associated video camera, so that a visualization as well as fusion-based processing can be achieved. An experiment in central Beijing is presented, demonstrating that a large quantity of physical dimension and detailed traffic data can be obtained through such a system.

This research aims at generating an omnidirectional perception at the host vehicle's surroundings, extracting accurate and continuous motion trajectories of the nearby vehicles using low cost laser scanners. A system of detecting and tracking on-road vehicles using multiple laser scanners is developed, where focuses are cast on solving data association of simultaneous measurements from multiple sensors at different viewpoints, and state estimation in case of partial observations in dense dynamic situations. Experimental results in freeways in Beijing are presented, system efficiency is demonstrated, where motion trajectories describing driving behaviors such as overtaking, lane changing and other interactions between driving objects are captured. In addition, the accuracy in vehicle detection and tracking is examined using a reference vehicle with a ground truth GPS.

Two challenging issues for object detection are how to separate near objects and how to separate similar different-class objects. Learned that during human's vision perception, tokens moving or functioning in a similar manner are perceived as one unit, stated by the common fate principle, we propose a method to detect objects of multiple classes. Our method extends the Implicit Shape Model (ISM) to incorporate motion grouping results of object parts, and meets the challenges. Keypoint-based object parts are firstly detected and then grouped by the similarities of their corresponding trajectories which are traced by keypoint tracking. The grouping results are combined into a Hough transform framework. In Hough transform based methods, each object part votes for object centers and labels according to a trained codebook. In our method, the votes are assigned different weights according to the motion grouping results. One vote is assigned larger weight if it has larger consistence with the votes of other object parts in the same motion group. In such a manner, peaks in the formed Hough images which correspond to object hypotheses become easier to find. And our method gains improvement in both object position and label estimation. Experiments are provided to show the merit in terms of detection accuracy.

Complex human activity detection is a challenging problem, especially when people interact with each other. Approaches utilizing local spatio-temporal features work well with background clutter, scale and illumination changing. However, most of them focus on classifying short video sequences. In real world applications such as surveillance, it's hard to get the well segmented video clip to classify. So how to detect and localize complex human activities in unsegmented videos is a problem need to be solved. In this paper, based on the local spatio-temporal feature, we propose a variation of Hough Voting method using the Implicit Shape Model which can localize and recognize complex human activity simultaneously. Our approach is tested on the UT-Interaction dataset, and demonstrates promising results in complex human activity detection and localization.

In the context of driving assistance, detection of moving targets is challenging when the automobile is witnessing very complex environment. The main challenges come from difficulties to tell foreground from moving background due to camera motion. We present an efficient method for target detection by hierarchically distinguishing the motion of targets from that of the background and using color information to help with the process. Firstly, point features are extracted and tracked to form feature trajectories. Secondly, distance measures are defined on the trajectories and hierarchically used to cluster the generated trajectories into regions. The last step is devoted to better detection of the targets by adding appearance information. Experiments show our method is able to give target detection results in real time even in very complex environments.

Motivated by two potential applications, i.e. enhancing driving safety and traffic data collection, a system has been developed using a single-layer horizontal laser scanner as the major sensor for both localization and perception of the surroundings in a large dynamic urban environment. This research focuses on a classification method, that given a stream of laser measurements, classify the moving object into either a person, a group of people, a bicycle or a car. In this research, a number of features are defined after examining the property of data appearance. A classification method is proposed after examining the likelihood measures between each pair of feature and class. Experimental results are presented, demonstrating that the algorithm has efficiency with respect to both driving safety and traffic data collection in highly dynamic environment.

We present a promising system to simultaneously detect and track multiple humans in the outside scene using laser and vision. The useful information of laser and vision is automatically extracted and combined in a Bayesian formulation. In order to compute MAP estimation, an effective probabilistic detection-based particle filter (PD-PF) has been proposed. Experiments and evaluations demonstrate that not only can our system perform robustly in real environments, but also obtain better approximation of MAP than previous methods in most complex situations.

This research is motivated by two potential applications - enhancing driving safety and collecting traffic data in a large dynamic urban environment. A laser scanner based approach is proposed, in which SLAM (simultaneous localization and mapping) is developed with moving object detection and tracking using a laser scanner for perception, using GPS to achieve global accuracy, and using yaw rate and wheel speed to diagnose pose errors. Experiments are conducted to collect data along a course (4.5 km) with a test-bed vehicle run in a highly dynamic environment. The algorithms are examined, possibilities with respect to the two potential applications are demonstrated, and future works are discussed.

In this research, we propose a method of SLAM in a dynamic large outdoor environment using a laser scanner. Focus are cast on solving two major problems: 1) achieving global accuracy especially in non-cyclical environment, 2) tackling a mixture of data from both dynamic and static objects. Algorithms are developed, where GPS data and control inputs are used to diagnose pose error and guide to achieve a global accuracy; Classification of laser points and objects are conducted not in an independent module but across the processing in a framework of SLAM with moving object detection and tracking. Experiments are conducted using the data from two test-bed vehicles, and performance of the algorithms are demonstrated.

Laser based people tracking systems have been developed for mobile robotic or intelligent surveillance areas. Existing systems rely on laser point clustering to extract object locations. However, in a crowded environment, laser points of different objects are often interlaced and undistinguishable and can not provide reliable features. This paper presents a novel and robust laser-based tracking method for people in crowds. Firstly, we propose a stable feature extraction method based on accumulated distribution of successive laser frames. Then a robust tracking filter is proposed based on the combination of independent Bayesian filter and sampling based data association filter. Evaluations with real data show that the proposed method is robust and effective. It achieves a significant improvement compared with existing trackers.

In this paper, we present evaluation of the performance of a laser-based multi-people detection and tracking system. We propose metrics for frame-based as well as object- based evaluation paradigms considering complexity in multi- target tracking. In frame-based evaluation, accuracy of people localization is tested without considering people identity. In object-based evaluation, identity is included, so that the cases of broken/merged/crossed trackers are evaluated. Finally, we applied these metrics to the tracking results on real data.