Guisheng Chen

Mining regular routes from GPS data for ridesharing recommendations

The widely use of GPS-enabled devices has provided us amount of trajectories related to individuals activities. This gives us an opportunity to learn more about the users daily lives and offer optimized suggestions to improve peoples trip styles. In this paper, we mine regular routes from a users historical trajectory dataset, and provide ridesharing recommendations to a group of users who share similar routes. Here, regular route means a complete route where a user may frequently pass through approximately in the same time of day. In this paper, we first divide users GPS data into individual routes, and a group of routes which occurred in a similar time of day are grouped together by a sliding time window. A frequency-based regular route mining algorithm is proposed, which is robust to slight disturbances in trajectory data. A feature of Fixed Stop Rate (FSR) is used to distinguish the different types of transport modes. Finally, based on the mined regular routes and transport modes, a grid-based route table is constructed for a quick ride matching. We evaluate our method using a large GPS dataset collected by 178 users over a period of four years. The experiment results demonstrate that the proposed method can effectively extract the regular routes and generate rideshare plan among users. This work may help ridesharing to become more efficient and convenient.

Web service discovery based on keyword clustering and ontology

One of the challenging problem that Web service technology is now facing is effective service discovery. To solve the deficiencies of Web service description, matching and choosing under WSDL language, this paper presents a web service discovery method based on keyword clustering and concept expansion, mainly from the content of Web service, reasoning of service request and service matching, through classification and subsumption of concept, this paper retrieve and extract Web service content. Meanwhile, use weighted bipartite graph to match user request and published Web service, so as to enhance Web service discovery ability.

Network thinking and network intelligence

Networks interact with one another and are recursive. Network intelligence and networked intelligence, as the way of knowledge representation, have become very active recently. What topological measures can be used to characterize properties of networks?What properties do different structures of real-world networks share, and why? How did theses properties come about? How do these properties affect the dynamics of such networks? How to use network topology to extend other dimensions? Given a real-world network with certain properties, what are the best ways to search for particular nodes? Furthermore, some specific implementations and examples of network intelligence will be given in this paper. Such as mining typical topologies, discovering sensitive links and important communities from real complex networks, networked control, and making a virtual reality of emergence phenomenon in complex systems.

Mining Frequent Composite Service Patterns

Service mining is an important technology which could help service composition. This paper mainly studied the problem of mining frequent composite services from the log of the composition-oriented service discovery system. First, a composite service representation method based on n-tuple was proposed. Second, a frequent composite service mining algorithm was put forward. Finally, an experiment was given to show the effectiveness of the mining algorithm.

A scaled-down traffic system based on autonomous vehicles: A new experimental system for ITS research

In this paper, we present our recent efforts on developing a physical environment for performing traffic experiments. The two main characteristics of this environment are: (1) the whole environment is scaled down from the real traffic, (2) the traffic behaviors are performed by numbers of miniature autonomous vehicles. Performing traffic experiments in an actual environment has long been a hard problem. Moreover, modeling traffic phenomena is also a tough task, due to the complexity of the natural traffic system. However, with the rapid development of autonomous vehicle technology, we have an opportunity to improve these problems from a new perspective. That is using autonomous vehicles to perform traffic experiments. But with the limitations in cost and land availability, directly using full size autonomous vehicles also seemed unrealistic. Thus, we built a 1/10 scaled-down traffic system (SDTS) with more than 50 miniature autonomous vehicles. The environmental design, autonomous vehicles developing, agent modeling, traffic control, and real-time monitoring is considered systematically during system design. The SDTS can be used as a repeatable, appraisable, and verifiable experimental platform for traffic researches, such as testing traffic solutions, verifying key technologies in intelligent vehicles, and performing experiments about ITS. By now, the SDTS has supported a series of workshops, exchange activities and competitions in China.

A decision-making method for unmanned cars based on drivable area cutting

Unmanned cars developing involves at least three major sections: environment perception, decision-making and subsequent vehicle control. Whereas the first and the latter have proven almost-solved problems, decision-making is considered to be the most important and difficult section in unmanned cars. In fact, no matter normal or unmanned cars, to ensure their safety, appropriate decision must be made timely by either human driver or computer. In this article, we present a decision-making method based on drivable area cutting. We considered the most common elements in actual driving process including traffic signs, traffic lights, pedestrians, vehicles and even other vehicles lights and horns to cut drivable area. We name it safe area for this cut-drivable area. Then decision is made based on the safe area. Experiment results show that our method works well in the highway and can be used in different environment with little change.

Spirit: A lightweight programming framework for intelligent system

This paper presents Spirit, a lightweight programming framework for intelligent system. Ease of use and tiny development effort are two key features considered during the design. Spirit is mainly a template class, which provides a configure table, commonly used libraries in system control and computer vision, interprocess communication for parallel programming, hardware middleware for different hardware products, and customizable algorithm functions. Developers utilize Spirit to create modules or even systems for their intelligent applications very quickly as well as using powerful tools in Spirit for convenient debugging. We compared Spirit with existing robot frameworks and proposed a new design criterion. Results show that Spirit gives a better experience on rapid system design and debugging. We used it when developing an intelligent vehicle system.

Multi-sensor information fusion for unmanned cars using radar map

Safety is the foremost quality to unmanned cars. In order to ensure the safety, unmanned cars must percept the surrounding environment precisely and exhaustively. To achieve this, various sensors including camera, lidar, and radar are equipped with unmanned cars. While the environment perception algorithms are relatively mature, there is no general solution for the multi-sensor information fusion for unmanned cars. In this article, we present a solution for multi-sensor information fusion for unmanned cars using radar map. With this solution, different environment information detected by various sensors can fuse naturally in radar map. Besides, the radar map is essentially a matrix and can be easily stored in memory. Experiment results show that radar map works well in all road conditions. And software development practices for unmanned cars also show that radar map can provide well support to decision-making, path planning and other subsequent sections.

A variable step-size LMS algorithm based on cloud model with application to multiuser interference cancellation

This paper presents a variable step-size Least Mean Square (LMS) algorithm based on cloud model which is a new cognitive model for uncertain transformation between linguistic concepts and quantitative values. In this algorithm, we use the error differences between two adjacent iteration periods to qualitatively estimate the state of the algorithm, and translate it into a propriety step-size in number according to the linguistic description of basic principle of variable step-size LMS. Simulation results show that the proposed algorithm is able to improve the steady-state performance while keeping a better convergence rate. We also apply this new algorithm to the multiuser interference cancellation, and results are also satisfied.