Academia.eduAcademia.edu

Figure 7 – uploaded by Lohith Kumar

See full PDFdownloadDownload figure
Fig. 7. Data set description analysis based on 52 transit surveillance-related papers surveyed in this paper. “None” refers to papers that include no reference to the data set used. “Complete” indicates that a full description is included, i.e., quantity and pixel resolution for both training and testing data. “Incom- plete” indicates that there is some description but not enough to account for “Complete.”

Figure 7 Data set description analysis based on 52 transit surveillance-related papers surveyed in this paper. “None” refers to papers that include no reference to the data set used. “Complete” indicates that a full description is included, i.e., quantity and pixel resolution for both training and testing data. “Incom- plete” indicates that there is some description but not enough to account for “Complete.”

Related Figures (11)

RELATED LITERATURE SURVEY SUMMARY
RELATED LITERATURE SURVEY SUMMARY
Fig. 1. Paper organization flowchart.
Fig. 1. Paper organization flowchart.
Fig. 2. Sample single-person or no interaction behavior. Suspicious person (marked with an ellipse) loitering for a long period of time without leaving in a bus.!
Fig. 2. Sample single-person or no interaction behavior. Suspicious person (marked with an ellipse) loitering for a long period of time without leaving in a bus.!
Fig. 3. Sample multiple-person interaction behavior. Pedestrians on a crosswalk.”
Fig. 3. Sample multiple-person interaction behavior. Pedestrians on a crosswalk.”
Fig. 4. Sample person—vehicle interaction. Person being run over by < vehicle. Fig. 5. Sample person-facility/location interaction. Object left behind in a train station.*
Fig. 4. Sample person—vehicle interaction. Person being run over by < vehicle. Fig. 5. Sample person-facility/location interaction. Object left behind in a train station.*
Fig. 6. Increasing interest on human behavior recognition research is shown through a comparison of a number of papers published up to 2008.
Fig. 6. Increasing interest on human behavior recognition research is shown through a comparison of a number of papers published up to 2008.
BEHAVIOR RECOGNITION SUMMARY ADVERTISED BY COMMERCIAL PROVIDERS IN THEIR WEBSITES using image-processing algorithms are created to be used on aerial platforms, tracking-based methods often used in current transit applications will likely have problems with aerial video. Tracking systems have problems with objects following broken trajectories resulting from limited field of view and occlusion due to terrain features. Recent work is being driven by these problems, leading to solutions for problems such as the study of global motion patterns [199] from aerial video. As resolution and video quality increases, transit surveillance, including peo- ple, vehicles, and behavior analysis, is logically the next step. interact to form a “bigger picture,” which can potentially offer a viable solution to complex problems. Many transit systems have large sensor networks (e.g., audio, video, motion sensors, and smoke detectors) already in place. Under such scenar- ios, multiple sensors can be used to generate more accurate, complete, and dependable data. For example, camera networks can be used to provide multiple views of a scene, which might diminish the number of tracking occlusions [184]. In addition, sensors can often overcome weaknesses of other sensors; for example, fusing color and infrared video can be used to improve tracking through occlusions [185]. However, there is not much work reported on the integration of different types of sensors in automated video surveillance systems [5]. Multimodal fusion, such as audio and video [186] or infrared and stereovision [187], can potentially offer better scene understanding, thereby improving situational awareness and response time. For general distributed surveillance, see a detailed survey in [5] for more information.
BEHAVIOR RECOGNITION SUMMARY ADVERTISED BY COMMERCIAL PROVIDERS IN THEIR WEBSITES using image-processing algorithms are created to be used on aerial platforms, tracking-based methods often used in current transit applications will likely have problems with aerial video. Tracking systems have problems with objects following broken trajectories resulting from limited field of view and occlusion due to terrain features. Recent work is being driven by these problems, leading to solutions for problems such as the study of global motion patterns [199] from aerial video. As resolution and video quality increases, transit surveillance, including peo- ple, vehicles, and behavior analysis, is logically the next step. interact to form a “bigger picture,” which can potentially offer a viable solution to complex problems. Many transit systems have large sensor networks (e.g., audio, video, motion sensors, and smoke detectors) already in place. Under such scenar- ios, multiple sensors can be used to generate more accurate, complete, and dependable data. For example, camera networks can be used to provide multiple views of a scene, which might diminish the number of tracking occlusions [184]. In addition, sensors can often overcome weaknesses of other sensors; for example, fusing color and infrared video can be used to improve tracking through occlusions [185]. However, there is not much work reported on the integration of different types of sensors in automated video surveillance systems [5]. Multimodal fusion, such as audio and video [186] or infrared and stereovision [187], can potentially offer better scene understanding, thereby improving situational awareness and response time. For general distributed surveillance, see a detailed survey in [5] for more information.
Related topics:
Civil EngineeringImage ProcessingTerrorismTransportationFacility LocationAnomaly DetectionEvent DetectionMotion estimationMotion DetectionMoving Object Recognition
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved