Which is a more accurate estimation of SOC between different versions of Kalman filter vs advanced Fuel gauge ICs such as the ones from Texas Instruments (Impedance Track algorithm) or Maxim Integrated (Model Guage algorithm)?
Using a Kalman filter alone would most likely not be that accurate unless a lot of battery specific algorithms are added. The Maxim and TI solutions implement a lot of algorithms that model specific battery behaviors with regard to temperature dependence, aging, and charge/discharge rates. It would be very difficult to match their performance over the wide range of conditions batteries are subjected to.
Besides accuracy, there are several other advantages of using a specific fuel gauge ICs. The power consumption of these fuel gauge ICs is much lower than a typical general purpose microcontroller. This is very important for applications where a lot of time is spent in sleep mode because you don't want the fuel gauging algorithm to be waking up the MCU and become a significant consumer of power.
The fuel gauges also provide a lot of additional features such as temperature, overcurrent, overvoltage alerts and also calculate parameters such as Time-to-empty (TTE), time-to-full (TTF), and capacity. These may or may not be important for your application.
Full Disclosure: I work for Maxim on their fuel gauges. If you were to ask me, I would recommend the MAX17055 as a good choice for many applications. It implements a feature called Modelgauge m5 EZ which allows designs to be done without battery testing and characterization. Battery characterization is often required for many other ICs, and it can take a few weeks of test time. Using one of these ICs is significantly easier than writing the algorithms from scratch!