Fully automated video based pedestrian detection and tracking is a challenging task with many practical and important applications. We present our work aimed to allow robust and simultaneously close to real-time tracking of pedestrians. The presented approach is stable to occlusions, lighting conditions and is generalized to be applied on arbitrary video data. The core tracking approach is built upon tracking-by-detections principle. We describe our cascaded HOG detector with successive CNN verification in detail. For the tracking and re-identification task, we did an extensive analysis of appearance based features as well as their combinations. The tracker was tested on many hours of video data for different scenarios; the results are presented and discussed.
Reliable automatic detection of traffic jam occurrences is of big significance for traffic flow analysis related applications. We present our work aimed at the application of video based real-time traffic jam detection. Our method can handle both calibrated and un-calibrated scenarios, operating in world and in image coordinate systems respectively. The method is designed to be operated on a smart camera, but is also suitable for a standard personal computer. The combination of state-of-the-art algorithms for vehicle detections and velocity estimation allows robust long-term system operation in due to the high recall rate and very low false alarm rate. The proposed method not only detects traffic jam events in real-time, but also precisely localizes traffic jams by their start and end positions per road lane. We describe also our strategy in making computationally heavy algorithms real-time capable even on hardware with a limited computing power.
This paper presents our work towards robust vehicle detection in dynamic and static scenes from a brief historical perspective up to our current state-of-the-art. We cover several methods (PCA, basic HOG, texture analysis, 3D measurement) which have been developed for, tested, and used in real-world scenarios. The second part of this work presents a new HOG cascade training algorithm which is based on evolutionary optimization principles: HOG features for a low stage count cascade are learned using genetic feature selection methods. We show that with this approach it is possible to create a HOG cascade which has comparable performance to an AdaBoost trained cascade, but is much faster to evaluate.
The efficient and robust detection of the presence of vehicles in restricted parking areas is important for applications in law enforcement as well as for the enforcement of parking rules on private property. We present our work towards this goal aimed at the application of vehicle detection in urban environments. The method is to be suited for smart cameras which have to operate autonomously over extended periods of time. Our system is developed as part of a bigger research effort which combines onsite vehicle presence detection and an associated web management system which is intended to monitor, steer and reroute delivery vehicles.
This paper applies object detection in a microscopic traffic model calibration process and analyses the outcome. To cover
a large and versatile amount of real world data for calibration and validation processes this paper proposes semiautomated
data acquisition by video analysis. This work concentrates mainly on the aspects of a automatic annotation
tool applied to create trajectories of traffic participants over space and time.
The acquired data is analyzed with a view towards calibrating vehicle models, which navigate through a road's surface
and interact with the environment. The applied vehicle tracking algorithms for automated data extraction provide many
trajectories not applicable for model calibration. Therefore, we applied an additional automated processing step to filter
out faulty trajectories. With this process chain, the trajectory data can be extracted from videos automatically in a quality
sufficient for the model calibration of speeds, the lateral positioning and vehicle interactions in a mixed traffic
This paper presents a vision based tracking system developed for very crowded situations like underground or railway
stations. Our system consists on two main parts - searching of people candidates in single frames, and tracking them
frame to frame over the scene. This paper concentrates mostly on the tracking part and describes its core components in
detail. These are trajectories predictions using KLT vectors or Kalman filter, adaptive active shape model adjusting and
texture matching. We show that combination of presented algorithms leads to robust people tracking even in complex
scenes with permanent occlusions.
The efficient monitoring of traffic flow as well as related surveillance and detection applications demand an increasingly robust recognition of vehicles in image and video data. This paper describes two different methods for vehicle detection in real world situations: Principal Component Analysis and the Histogram of Gradients principle. Both methods are described and their detection capabilities as well as advantages and disadvantages are compared. A large sample dataset which contains images of cars from the backside and frontside in day and night conditions is the basis for creating and optimizing both variants of the proposed algorithms. The resulting two detectors allow recognition of vehicles in frontal view +- 30 deg and views from behind +- 30 deg. The paper demonstrates that both detection methods can operate effectively even under difficult lighting situations with high detection rates and a low number of false positives.
This paper describes a close to real-time scale invariant implementation of a pedestrian detector system which is based on the Histogram of Oriented Gradients (HOG) principle. Salient HOG features are first selected from a manually created very large database of samples with an evolutionary optimization procedure that directly trains a polynomial Support Vector Machine (SVM). Real-time operation is achieved by a cascaded 2-step classifier which uses first a very fast linear SVM (with the same features as the polynomial SVM) to reject most of the irrelevant detections and then computes the decision function with a polynomial SVM on the remaining set of candidate detections. Scale invariance is achieved by running the detector of constant size on scaled versions of the original input images and by clustering the results over all resolutions. The pedestrian detection system has been implemented in two versions: i) fully body detection, and ii) upper body only detection. The latter is especially suited for very busy and crowded scenarios. On a state-of-the-art PC it is able to run at a frequency of 8 - 20 frames/sec.
A vision system designed to detect people in complex backgrounds is presented. The purpose of the proposed algorithms is to allow the identification and tracking of single persons under difficult conditions - in crowded places, under partial occlusion and in low resolution images. In order to detect people reliably, we combine different information channels from video streams. Most emphasis for the initialization of trajectories and the subsequent pedestrian recognition is placed on the detection of the head-shoulder contour. In the first step a simple and fast shape model selects promising candidates, then a local active shape model is matched against the gradients found in the image with the help of a cost function.
Texture analysis in the form of co-occurrence features ensures that shape candidates form coherent trajectories over time. In order to reduce the amount of false positives and to become more robust, a pattern analysis step based on Eigenimage analysis is presented.
The cues which form the basis of pedestrian detection are integrated into a tracking algorithm which uses the shape information for initial pedestrian detection and verification, propagates positions into new frames using local motion and matches pedestrians with the help of texture information.