TV Doorphone Project

The "TV doorphone" was my first project and Matsushita Denko (MEW) in the image processing group. MEW had a previous analog based product in which a small video camera where a doorbell would normally be placed. A small TV screen and phone allowed someone in the house to see and communicate with the visitor when the "doorbell" was pressed. In that previous product, the main customer complaint was a too narrow field of view and viewing problems under poor lighting conditions. Although the viewing angle could be widened by using a high quality glass lense, it was prohibitively expensive. Cheap plastic lenses could provide a relatively wide field of view but their optics produce the "fisheye" effect that is uncomfortable to viewers because of the distortion of the visitors faces. (Figure 1)

The project I intiated was to develop a digital signal processor based TV doorphone where the lense distortion could be corrected in real time along with features such as lighting correction and digital zoom. Together with reserachers in the Semiconductor Group who used this project as their model project for VHDL based ASIC design, a 40,000 gate CMOS ASIC and actual product was produced after 18 months.

One of the interesting aspects was the human perception aspects of the lens correction. Algorithms that were mathematically correct to correct in both the vertical and horizontal directions were often labelled as unnaturall by test viewers. (Figure 2) In the end, a variable algorithm that adjusted the fisheye distortion only in the horizontal direction was implemented. (Figure 3) This also had the added benefit of greatly reducing memory and processing gate requirements.

The "GenKanBan" product sold well for several years. (Advertisement in Figure 4) It also was selected as one of the best new products of 1993 by the Japan Electrical Manufacturing Association (JEMA). (Figure 5) Patents: USA: 5,815,199 Japan: 2658642 (Publication number: HEI 05-252513)

Article in Japan Electronic Engineering magazine (Japanese 5 Mbyte)

Figure 4

Figure 5