| Presentation | 2022-03-07 [Poster Presentation] Head location estimation using CSRNet for understanding Highly Congested Scenes Takuya Nagatoshi, Michiharu Niimi, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Neural network for congested scene recognition called CSRNet has been proposed. The purpose of CSRNet is to count the number of heads based on the output of CSRNet. We plane to apply CSRNet to the attendance management system which is able to recognize who are attending in a classroom, in order to do that, we need to estimate head location. In CSRNet, the input is an RGB image that is taken for a scene, and the output is the density map of people. The number of heads is given by the integral of density map. The learning process of CSRNet is performed by decreasing the mean square error between the output and the correct density map. Note that the size of density map is 1/8 of input image. Because the density map can be regarded as 2-dimentinal signal, we can estimate head locations by determining its maximal value. In this report, we put an up-sampling layer inside of CSRNet to try to make a detailed density map. The size of density map become 1/4 of input image. We apply gaussian filters to density map to reduce the influence of noise. In the experiments, we used ShanghaiTech data set and real pictures with is taken at a real classroom. As a result of experiments, we confirmed that the effectiveness of adding up-sampling layer, but it is difficult to adjust head location error and multiple extraction. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | congested scene recognition / head location estimation / CSRNet |
| Paper # | EMM2021-95 |
| Date of Issue | 2022-02-28 (EMM) |
| Conference Information | |
| Committee | EMM |
|---|---|
| Conference Date | 2022/3/7(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | (Primary: Online, Secondary: On-site) |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Image and Sound Quality, Metrics for Perception and Recognition, Human Auditory and Visual System, etc. |
| Chair | Ryoichi Nishimura(NICT) |
| Vice Chair | Masaaki Fujiyoshi(Tokyo Metropolitan Univ.) / Masatsugu Ichino(Univ. of Electro-Comm.) |
| Secretary | Masaaki Fujiyoshi(Utsunomiya Univ.) / Masatsugu Ichino(NICT) |
| Assistant | Shoko Imaizumi(Chiba Univ.) / Youichi Takashima(Kaishi Professional Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Enriched MultiMedia |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | [Poster Presentation] Head location estimation using CSRNet for understanding Highly Congested Scenes |
| Sub Title (in English) | |
| Keyword(1) | congested scene recognition |
| Keyword(2) | head location estimation |
| Keyword(3) | CSRNet |
| 1st Author's Name | Takuya Nagatoshi |
| 1st Author's Affiliation | Kyushu Institute of Technology(KIT) |
| 2nd Author's Name | Michiharu Niimi |
| 2nd Author's Affiliation | Kyushu Institute of Technology(KIT) |
| Date | 2022-03-07 |
| Paper # | EMM2021-95 |
| Volume (vol) | vol.121 |
| Number (no) | EMM-417 |
| Page | pp.pp.17-22(EMM), |
| #Pages | 6 |
| Date of Issue | 2022-02-28 (EMM) |