In recent years, family fitness has been widely concerned. In order to get a truly convenient, efficient and immersive fitness experience in the family scene, AI fitness cameras and large TV screens are indispensable. With the diversification of family fitness needs and the continuous emergence of various intelligent terminals, intelligent cameras that can intelligently empower family fitness scenes are also becoming the best choice for family fitness enthusiasts. Therefore, this paper designs an AI fitness camera to meet the needs of users for family fitness.
1 Overall system scheme
The overall system scheme collects sports information of fitness enthusiasts through Gekewei GC4C33, and then inputs the information to Ruixin Micro RK3566, which integrates the collected sports information of fitness enthusiasts with sports fitness apps and AI algorithms, and finally outputs it to the TV screen through HDMI. The system adopts Type-C power supply and has audio interface. The overall system scheme is as follows:
2 Hardware system design
The main control of hardware adopts Ruixin Micro RK3566 scheme, and the sensor uses Geke Micro GC4C33. Ruixin Micro RK3566 chip is a general-purpose SoC designed for consumer industry applications. CPU adopts 4-core A55 architecture processor, integrated G52 graphics processor, built-in independent NPU, running Android 11. The RK3566 has a very powerful video decoding capability. It supports high-definition decoding of 4K H.264/H.265/VP9 and other formats, and simultaneous decoding of multiple video sources, such as 8-10 channels of 1 080P H264/H265. The RK3566 supports HDR10, with excellent performance in color and dynamic range, and supports image post-processing, deinterleaving, denoising, color enhancement, and super-resolution. The RK3566 adopts a new independent JPEG decoding processor, which can efficiently and concurrently process multiple small image parsing. It supports H.264 and H.265 encoding at 1080P/60 fps, and supports dynamic code rate, frame rate, resolution adjustment and other functions. RK3566 has a built-in high-performance ISP, which can provide 8 M@30 Fps processing capability, time-sharing multiplexing to meet the dual camera needs. HDR function is supported to make the image clear under backlight or strong light. Support dual channel simultaneous zoom output function, support noise elimination function, so that the image under weak light conditions is also delicate. The RK3566 supports the defogging function, which can be seen clearly in foggy days. It supports horizontal correction of LDCH to remove the distortion caused by the sensor lens. The RK3566 supports the simultaneous input of two CSI (2Lane)+one DVP interface cameras at most.
GC4C33 is a high-quality 4 megapixel CMOS image sensor, which is suitable for camera products, AI fitness, digital camera products and mobile camera applications. GC4C33 includes one 2 560 H × 1 440 V pixel array, on-chip 12/10 bit ADC and image signal processor. The comprehensive integration of high performance makes GC4C33 meet the design requirements and reduces the implementation process. It provides RAW12 and RAW10 data formats with MIPI interfaces. There is a common two-wire serial interface for the host to control the operation of the entire sensor.
The quality of PCB design directly affects the success or failure of product development, and the mipi signal design of power supply, clock and sensor is the most important. AVDD_ The PLL and the DVDD core power supply are isolated by magnetic beads. A magnetic bead is connected in series between the ground filter capacitor and the ground. The connected magnetic beads are placed as close to the power supply pins as possible. At least two ground vias are left at the ground terminal. AVDD33_ The PLL is isolated from the digital 3.3 V power supply by magnetic beads. AVDD33_ DDR_ The PLL is isolated from the digital 3.3 V power supply by magnetic beads. XIN, XOUT, RTC of crystal_ XIN, RTC_ The routing of XOUT signals is processed in a package way throughout the whole process, and the design ensures that these signals have a complete reference. No high-speed signal can pass under the crystal oscillator. When designing MIPI signals, the differential signals take GND as the reference plane, and keep the reference integrity. The differential signals need to be processed in packages; The PCB wiring is less than 4 inches, the differential pair P/N equal length is controlled within ± 5 mils, the sampling differential clock is used as reference between pairs, and the equal length is controlled within ± 300 mils. The PCB routing control differential impedance of MIPI RX differential pair is 100 Ω± 10%; When the differential signal passes through the connector, the GND pin must be used for isolation between adjacent differential signal pairs.
3 Conclusion
The hardware of this paper adopts the Ruixin micro RK3566 scheme, and the software uses the Android system, developed with JAVA programming language, which can be transplanted and reused across platforms, and can be compatible with applications developed by third parties in the Android system. The experimental results show that the AI fitness camera designed in this scheme is simple and easy to use, and has been well received by fitness enthusiasts.