News Release

Video camera in a public place knows the density of people or vehicle more accurately

Peer-Reviewed Publication

Japan Advanced Institute of Science and Technology

Figure 1

image: Figure: An example of density map obtained from an image in TRANCOS dataset. view more 

Credit: JAIST

Deep learning applied for image/video processing opened the door for the practical deployment for object detection and identification with acceptable accuracy. Crowd counting is another application of image/video processing. The scientists at Japan Advanced Institute of Science and Technology (JAIST) designed a new DNN with backward connection, which achieved more accurate estimation of the density of objects. It can be applied for estimating human density in the public or vehicle density on a road in order for improving public safety/security and traffic efficiency.

Video surveillance is one of the standard ways for obtaining information to detect the status of objects. For example, video surveillance employed on a road is monitored to obtain the information on the flow of traffic, occurrence of accident, and/or the density of the vehicle for the purpose of improving security, safety, and/or efficiency of traffic. Another example of video surveillance is human traffic in the public. Monitoring the flow and the density of the people is mandatory to assure the safety of public places, especially for indoor environment.

Obtaining the information of the density or the number of objects, such as vehicles or people, is called crowd counting. Crowd counting with higher accuracy will offer more seamless control of ITS with less 'jaggy' feedback, or will detect serious status of human congestion that may cause accidents properly. The research group in JAIST led by Dr. Sooksatra and Prof. Atsuo Yoshitaka in collaboration with a research group of SIIT in Thailand proposed a new network employing backward connections in DNN, which achieved higher performance in crowed computing.

"Backward connection in DNN enables to take advantages obtained from both high-level feature and low-level feature in an image, and therefore achieves higher performance than before" says Prof. Atsuo Yoshitaka, the head of Yoshitaka Lab. The Yoshitaka lab. is currently developing different kind of DNNs for industrial applications such as object detection/identification of objects in micrograph, defect detection for industrial products, and DNA analysis for automated diagnosis.


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.