Ai On Raspberry Pi | AI Final Year Project Report

Complete Technical Paper Content — Ai On Raspberry Pi

1. Executive Abstract, Problem Formulation & Background

Raspberry Pi ( PY) is a series of small single-board computers (SBCs) originally developed in the United Kingdom by the Raspberry Pi Foundation in collaboration with Broadcom. To commercialize the product and support its growing demand, the Foundation established a commercial entity, now known as Raspberry Pi Holdings. The Raspberry Pi was originally created to help teach computer science in schools, but gained popularity for many other uses for its low cost, high performance, compact size, and flexibility. It is now used in areas such as industrial automation, robotics, home automation, IoT devices, and hobbyist projects. The company's products range from simple microcontrollers to computers that the company markets as being powerful enough to be used as a general purpose PC. Computers are built around a custom designed system on a chip and offer features such as HDMI video/audio output, USB ports, wireless networking, GPIO pins, and up to 16 GB of RAM. Storage is typically provided via microSD cards. In 2015, the Raspberry Pi surpassed the ZX Spectrum as the best-selling British computer of all time. As of March 2026, over 75 million units had been sold.

The first prototypes resembled small USB sticks. By August 2011, fifty functionally complete "alpha" boards were produced for testing, with demonstrations showing them running a Debian-based desktop and handling 1080p video playback. In late 2011, twenty-five "beta" boards were finalized, and to generate publicity before the official launch, ten of these were auctioned on eBay in early 2012. The first commercial Raspberry Pi, the Model B, was launched on 29 February 2012, with an initial price of $35. Demand far exceeded expectations, causing the websites of the two initial licensed distributors, Premier Farnell and RS Components, to crash from high traffic. Initial batches sold out almost immediately, with one distributor reporting over 100,000 pre-orders on the first day. The lower-cost $25 Model A followed on 4 February 2013. The Raspberry Pi did not ship with a pre-installed operating system. While ports of RISC OS 5 and Fedora Linux were available, a port of Debian called Raspbian quickly became the standard. Released in July 2012, it was optimized to leverage the Raspberry Pi's floating-point unit, offering significant performance gains. Raspberry Pi quickly endorsed it as the official recommended OS, and by September 2013, the company assumed leadership of Raspbian's development.

The flagship Raspberry Pi series offers high-performance hardware, a full Linux operating system, and a variety of common ports in a compact form factor roughly the size of a credit card.

The Model B (2012) features a 700 MHz single-core 32-bit ARM11 CPU, a VideoCore IV GPU, 256 then 512 MB RAM and a 26-pin GPIO header. The Model A (2013) is a smaller, lower-cost version with 256 MB RAM, no Ethernet, and fewer USB ports. The Model B+ and Model A+ (2014) add a 40-pin GPIO header, microSD card support, and replace the RCA video connector with a combined 3.5 mm audio/video jack. The Raspberry Pi 2 Model B v 1.1 (2015) includes a 900 MHz quad-core Cortex-A7 CPU and 1 GB of RAM. The Raspberry Pi 2 Model B, v 1.2 (2016) and v 1.3 (2025) include a 900 MHz quad-core Cortex-A53 CPU and 1 GB of RAM. The Raspberry Pi 3 Model B (2016) features a 1.2 GHz quad-core 64-bit Cortex-A53 CPU, 1 GB RAM, Wi-Fi, Bluetooth, and USB boot support. The Raspberry Pi 3 Model B+ (2018) upgrades to a 1.4 GHz CPU, 1 GB RAM, faster Ethernet, dual-band Wi-Fi, and Power over Ethernet (PoE) support. The Raspberry Pi 3 Model A+ (2018) is the final A-series model, offering the same features as the 3B+, but with 512 MB RAM and in a smaller form factor. The Raspberry Pi 4 (2019) introduces a 1.5 GHz quad-core Cortex-A72 CPU, a VideoCore VI GPU, USB 3.0 ports, true Gigabit Ethernet, support for dual 4K monitors, and options for 1, 2, 3, 4, or 8 GB of RAM. The Raspberry Pi 5 (2023) features a 2.4 GHz quad-core Cortex-A76 CPU, a VideoCore VII GPU, PCIe support, and options for 1, 2, 4, 8, or 16 GB of RAM.

The Keyboard Computer series combines Raspberry Pi board and a keyboard into a keyboard computer form factor, providing a self-contained Linux-based desktop system.

The Raspberry Pi 400 (2020) features a custom board based on the Pi 4. It includes a 1.8 GHz quad-core Cortex-A72 processor, 4 GB of RAM, and a large integrated heat sink. It supports dual 4K monitors via two micro HDMI ports and includes gigabit Ethernet. The Raspberry Pi 500 (2024) is based on the Pi 5 and succeeds the Pi 400. It features a 2.4 GHz quad-core Cortex-A76 processor and 8 GB of RAM. Unlike the Raspberry Pi 5, it lacks a PCIe interface. The Raspberry Pi 500+ (2025) is based on the Pi 5. It features the same 2.4 GHz quad-core Cortex-A76 processor as the Pi 500, 16 GB of RAM and an M.2 slot with a 256 GB SSD preinstalled. It replaces the membrane keyboard of the original model for a mechanical variant with RGB lighting.

2. Related Work, Literature Survey & Algorithmic Foundations

The Raspberry Pi Zero is a series of compact, low-cost, and low-power single-board computers that provide basic functionality and Linux compatibility for embedded and minimalist computing applications.

The Raspberry Pi Zero (2015), priced at US$5, features a 1 GHz single-core 32-bit ARM11 CPU, 512 MB of RAM, mini HDMI, and micro USB ports for data and power. It includes an unpopulated 40-pin GPIO header. The Zero v1.3 (2016) added a camera connector. The Zero W (2017) introduced onboard Wi-Fi and Bluetooth for US$10. The Zero WH (2018) added pre-soldered GPIO pins for US$15. The Raspberry Pi Zero 2 W (2021), priced at US$15, features a quad-core 64-bit ARM Cortex-A53 CPU and includes wireless connectivity. The Zero 2 WH variant adds a pre-soldered GPIO header for US$18.

The Pico series consists of compact microcontroller boards based on Raspberry Pi-designed chips. Unlike other models, Pico boards do not run Linux or use removable storage. Instead, programs are loaded by flashing binaries to onboard flash memory. The boards have GPIO on castellated edges for soldering directly to carrier boards, and they are spaced to be compatible with standard breadboards. Versions with pre-soldered header pins mounted on the underside are available for an additional US$1.

The Raspberry Pi Pico (2021) was the first board based on the in-house RP2040 microcontroller. It features a dual-core 32-bit ARM Cortex-M0+ CPU, 264 KB of RAM, and 2 MB of flash memory, and was launched at US$4. The Pico W (2022) adds Wi-Fi and Bluetooth and was launched at US$6. The Raspberry Pi Pico 2 (2024) introduced the RP2350 microcontroller, featuring a pair of 32-bit ARM Cortex-M33 cores and a pair of RISC-V cores (the user can select either), 520 KB of RAM, and 4 MB of flash memory, priced at US$5. The Pico 2 W adds Wi-Fi and Bluetooth at US$7.

The Compute Module (CM) series delivers Raspberry Pi's flagship hardware in a compact form for industrial and embedded applications, omitting onboard ports and GPIO headers in favour of a carrier board interface. Compute Modules are offered in one of two formats: a board matching the physical dimensions of a DDR2 SO-DIMM RAM module (though electrically incompatible with standard SO-DIMM sockets) and a smaller board with two 100-pin high-density connectors that enable additional interfaces.

Compute Module 1 (2014) – Based on the original Raspberry Pi. Features a single-core ARM11 CPU, 512 MB RAM, and 4 GB eMMC flash storage. SO-DIMM form factor. Compute Module 3 (2017) – Based on the Pi 3. Includes a quad-core 64-bit Cortex-A53 CPU, 1 GB RAM, and 4 GB eMMC; also available as a "Lite" variant without eMMC. SO-DIMM form factor. Compute Module 3+ (2019) – Based on the Pi 3+. Offers 0 (Lite), 8, 16, or 32 GB eMMC options. SO-DIMM form factor. Compute Module 4 (2020) – Based on the Pi 4. Includes a quad-core 64-bit Cortex-A72 CPU, 1, 2, 4, or 8 GB RAM, and 0 (Lite), 8, 16, or 32 GB eMMC; optional Wi-Fi and Bluetooth. High-density connector form factor; CM4S variant uses SO-DIMM form factor. Compute Module 5 (2024) – Based on the Pi 5. Features a quad-core 64-bit Cortex-A76 CPU, 2, 4, 8, or 16 GB RAM, and 0 (Lite), 16, 32, or 64 GB eMMC; optional Wi-Fi and Bluetooth. High-density connector form factor.

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

Storage is typically provided via a microSD card, though some Compute Modules offer onboard eMMC flash. Newer models support booting from USB storage, and the Pi 5 includes support for NVMe SSDs over PCIe. The original Model A and Model B use a full-sized SD card slot. Boards also include USB ports for peripherals such as keyboards, mice, and storage devices.

Most Raspberry Pi models include a 40-pin connector known as the GPIO (general-purpose input/output) header, although only some of the pins are dedicated to GPIO functions. The header, designated as J8, uses a consistent pinout across models. The header supplies 3.3 V and 5 V power along with various multiplexed, low-speed interfaces, including UART, SPI, I²C, I²S, and PCM. GPIO pins can be configured as either inputs or outputs. When set as an output, a pin can drive a high (3.3 V) or low (0 V) signal. When configured as an input, it can read a high (3.3 V) or low (0 V) voltage level. The original Raspberry Pi 1 Model A and B include only the first 26 pins of this header. On some Pi Zero models, the header is unpopulated, but solderable through-holes are provided. The Pico models feature a unique layout with unpopulated through-holes and a castellated edge, allowing it to be surface-mounted as a module. Compute Module boards do not include GPIO headers but instead expose GPIO signals through their board connectors.

The recommended operating system is Raspberry Pi OS, a Debian-based Linux distribution optimized for Raspberry Pi hardware and tuned to have low base memory requirements. It is available in both 32-bit and 64-bit versions and comes in several editions: a standard edition, a "Lite" version without a desktop environment, and a "Full" version that includes a comprehensive suite of software. Raspberry Pi OS can be purchased pre-installed on a microSD card, or downloaded and installed using Raspberry Pi Imager, a utility introduced in March 2020 to simplify the installation of operating systems onto SD cards and other media for Raspberry Pi devices. Available for macOS, Raspberry Pi OS, Ubuntu, and Microsoft Windows, Imager allows users to download and write operating system disk images within a single application. In addition to Raspberry Pi OS, the utility supports a variety of third-party operating systems, including Alpine Linux, Arch Linux ARM, Armbian, Emteria.OS (Android based), FreedomBox, Kali Linux, LibreELEC, RetroPie, RISC OS, SatNOGS, Ubuntu, and Windows.

Raspberry Pi systems use Broadcom's VideoCore GPU, which requires a proprietary firmware binary blob to be loaded at boot. Initially, the supporting software stack was entirely proprietary, though parts of the code were later released. Most driver functionality remains within the proprietary GPU firmware, accessed via closed-source runtime libraries such as OpenMAX IL, OpenGL ES, and OpenVG. These libraries interface with a kernel-space open-source driver, which in turn communicates with the proprietary GPU firmware. Applications use OpenMAX IL for video, OpenGL ES for 3D graphics, and OpenVG for 2D graphics, with all graphics libraries making use of the closed-source EGL interface. In February 2020, Raspberry Pi announced the development of a Vulkan graphics driver. A working prototype demonstrated high performance in Quake III Arena on a Raspberry Pi 3B+ later that year. On 24 November 2020, Raspberry Pi 4's Vulkan driver was declared conformant.

Raspberry Pi offers several official camera modules that connect via the Camera Serial Interface. These modules are used for photography, video capture, and machine vision applications.

Camera Module (2013) – A 5-megapixel (MP) camera based on the OmniVision OV5647 sensor, supporting video resolutions up to 1080p. A version without an infrared filter (NoIR) was available for night-vision applications when used with infrared lighting. This model is no longer produced. Camera Module 2 (2016) – 8 MP Sony IMX219 sensor. Also available in NoIR version. High Quality Camera (2020) – 12.3 MP Sony IMX477 sensor. Supports interchangeable C/CS mount or M12 mount lenses and includes a tripod thread. Not available in a NoIR version, but IR filter can be removed. Camera Module 3 (2023) – 12 MP Sony IMX708 sensor with support for autofocus and high dynamic range. Offered in four variants: standard, wide field of view (FoV), NoIR, and NoIR wide FoV. Global Shutter Camera (2023) – 1.6 MP Sony IMX296 sensor with global shutter for high-speed imaging. Supports C/CS mount lenses and includes a tripod mount. Not available in a NoIR version, but IR filter can be removed. AI Camera (2024) – 12.3 MP Sony IMX500 sensor with integrated on-sensor processing capabilities for AI applications.

4. Neural Network Backbone, Layer Specifications & Loss Formulations

Raspberry Pi Touch Display (2015) – A 7-inch capacitive touchscreen. Raspberry Pi Touch Display 2 (2024) – A revised version of the original 7-inch touchscreen display with improved performance and compatibility. Also available in an 5-inch version since 2025. 10-inch version release in July 2026. Raspberry Pi Monitor (2024) – A 15.6-inch Full HD IPS display with built-in speakers and folding stand.

Official Raspberry Pi HATs (Hardware Attached on Top) and expansion boards extend the functionality of Raspberry Pi computers. The HAT standard was introduced in July 2014. Many boards use an EEPROM for automatic configuration.

AI HAT+ (2024) – A HAT for the Raspberry Pi 5 featuring a built-in Hailo-8L chip providing 13 TOPS of AI acceleration, or Hailo-8 chip providing 26 TOPS. AI HAT+ 2 (2026) – A HAT for the Raspberry Pi 5 featuring a built-in Hailo-10H chip providing 40 TOPS of AI acceleration, along with 8 GB of dedicated on-board RAM. Build HAT (2021) – Designed to interface with Lego Technic motors and sensors. Codec Zero – A compact audio input/output board for the Pi Zero. DAC+ / DAC Pro / DigiAMP+ – A range of audio HATs offering high-resolution digital-to-analogue conversion, with the DigiAMP+ including a built-in amplifier. M.2 HAT+ (2024) – A HAT for the Raspberry Pi 5 featuring an interface for attaching M.2 peripherals. AI Kit (2024) – A bundle including the M.2 HAT+ and a Hailo AI-8L module providing 13 TOPS of AI acceleration. SSD Kit (2024) – A bundle including the M.2 HAT+ and a NVMe SSD. Sense HAT (2015) – Includes sensors for temperature, humidity, pressure, orientation, and an 8×8 LED matrix with a joystick. Originally developed for the Astro Pi project. PoE+ HAT (2021) – Enables Power over Ethernet functionality for models with PoE support. TV HAT (2018) – Allows reception and decoding of digital DVB-T2 television broadcasts.

Technology writer Glyn Moody described the project in May 2011 as a "potential BBC Micro 2.0", not by replacing PC compatible machines but by supplementing them. In March 2012 Stephen Pritchard echoed the BBC Micro successor sentiment in ITPRO. Alex Hope, co-author of the Next Gen report, is hopeful that the computer will engage children with the excitement of programming. Co-author Ian Livingstone suggested that the BBC could be involved in building support for the device, possibly branding it as the BBC Nano. The Centre for Computing History strongly supports the Raspberry Pi project, feeling that it could "usher in a new era". Before release, the board was showcased by ARM's CEO Warren East at an event in Cambridge outlining Google's ideas to improve UK science and technology education. Harry Fairhead, however, suggests that more emphasis should be put on improving the educational software available on existing hardware, using tools such as MIT App Inventor to return programming to schools, rather than adding new hardware choices. Simon Rockman, writing in a ZDNet blog, was of the opinion that teens will have "better things to do", despite what happened in the 1980s. In October 2012, the Raspberry Pi won T3's Innovation of the Year award, and futurist Mark Pesce cited a (borrowed) Raspberry Pi as the inspiration for his ambient device project MooresCloud. In October 2012, the British Computer Society responded to the announcement of enhanced specifications by stating, "it's definitely something we'll want to sink our teeth into." In June 2017, Raspberry Pi won the Royal Academy of Engineering MacRobert Award. The citation for the award to the Raspberry Pi said it was "for its inexpensive credit card-sized microcomputers, which are redefining how people engage with computing, inspiring students to learn coding and computer science and providing innovative control solutions for industry." Clusters of hundreds of Raspberry Pis have been used for testing programs destined for supercomputers.

In June 2014, Polish industrial automation manufacturer TECHBASE released ModBerry, an industrial computer based on the Raspberry Pi Compute Module. The device has a number of interfaces, most notably RS-485/232 serial ports, digital and analogue inputs/outputs, CAN and economical 1-Wire buses, all of which are widely used in the automation industry. The design allows the use of the Compute Module in harsh industrial environments, leading to the conclusion that the Raspberry Pi is no longer limited to home and science projects, but can be widely used as an Industrial IoT solution and achieve goals of Industry 4.0. In March 2018, SUSE announced commercial support for SUSE Linux Enterprise on the Raspberry Pi 3 Model B to support a number of undisclosed customers implementing industrial monitoring with the Raspberry Pi. In January 2021, TECHBASE announced a Raspberry Pi Compute Module 4 cluster for AI accelerator, routing and file server use. The device contains one or more standard Raspberry Pi Compute Module 4s in an industrial DIN rail housing, with some versions containing one or more Coral Edge tensor processing units.

Official website Raspberry Pi, Department of Computer Science and Technology, University of Cambridge Raspberry Pi Wiki, supported by the RPF The MagPi Magazine "Raspberry Pi pinout" – board GPIO pinout "Raspberry Pi component map" Archived 7 June 2019 at the Wayback Machine "RaspberryPi Boards: Hardware versions/revisions" ARM1176JZF-S (ARM11 CPU Core) Technical Reference Manual Archived 19 June 2020 at the Wayback Machine, ARM Ltd.

5. Training Protocol, Convergence Behavior & Experimental Evaluation

Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. It is also a field of research spanning computer science, engineering, and mathematics that develops and studies methods and software enabling machines to perceive their environment and use learning and intelligence to take actions that maximise their chances of achieving defined goals. Prominent applications of AI include advanced web search engines, chatbots, virtual assistants, autonomous vehicles, play and analysis in strategy games (e.g., chess and Go), and content generation (e.g., text, images, audio, and videos). The traditional goals of AI research include learning, reasoning, knowledge representation, planning, natural language processing, and perception, as well as support for robotics. To reach these goals, AI researchers use techniques including state space search and mathematical optimisation, formal logic, artificial neural networks, and methods based on statistics, operations research, and economics. AI also draws upon psychology, linguistics, philosophy, neuroscience, and other fields. Some companies, such as OpenAI, Google DeepMind, and Meta, aim to create artificial general intelligence (AGI)—AI that can complete nearly any cognitive task at least as well as a human. Artificial intelligence was founded as an academic discipline in 1956. The field went through multiple cycles of optimism throughout its history, followed by periods of disappointment and loss of funding, known as AI winters. Funding and interest increased substantially after 2012, when graphics processing units (GPUs) started being used to accelerate neural networks, and deep learning outperformed previous AI techniques. This growth accelerated further after 2017 with the transformer architecture. The 2020s AI boom was spurred on largely by advances in generative AI, which can create and modify media. In addition to AI safety and unintended consequences and harms from the use of AI, ethical concerns, AI's long-term effects, environmental effects, and potential existential risks have prompted discussions of AI regulation.

AI programs use knowledge to answer questions intelligently and make deductions about real-world facts. Formal knowledge representation and knowledge engineering use symbols to represent words, concepts and things in the world. A knowledge base is a body of knowledge represented in a form that can be used by a program. An ontology is the set of objects, relations, concepts, and properties used by a particular domain of knowledge. Formal knowledge has been studied extensively since the 1970s and researchers have developed formalisms for a wide variety of difficult domains. The symbolic approach has difficulty with several problems: the breadth of commonsense knowledge (the set of atomic facts the average person knows is enormous), the sub-symbolic form of most commonsense knowledge (much of what people know is not represented as "facts" or "statements" they can express verbally), and knowledge acquisition (the problem of obtaining knowledge for AI applications). Large language models (and some other AI programs developed since 2012) do not require explicit, symbolic knowledge. They acquire knowledge by being trained on the combined text of millions of books and billions of websites. Modern AI can also learn about a domain by running experiments (as when AlphaZero learns game strategy by playing against itself). Machine learning solves the problems of general knowledge, commonsense knowledge and knowledge acquisition, however, this approach has struggled with accurate recall and valid reasoning.

Unsupervised learning analyses a stream of data, finds patterns, and makes predictions without any other guidance. Supervised learning requires labelling the training data with the expected answers, and comes in two main varieties: classification (where the program must learn to predict what category the input belongs in) and regression (where the program must deduce a numeric function based on numeric input). Reinforcement learning is when the agent is rewarded for good responses and punished for bad ones. The agent learns to choose responses that are classified as "good". Transfer learning is when the knowledge gained from one problem is applied to a new problem. Deep learning is a type of machine learning that runs inputs through biologically inspired artificial neural networks for all of these types of learning. Computational learning theory can assess learners by computational complexity, sample complexity (how much data is required), or other notions of optimisation.

Affective computing is a field that comprises systems that recognise, interpret, process, or simulate human affect (feeling, emotion, and mood). For example, some virtual assistants are programmed to speak conversationally or even banter humorously; it makes them appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate human–computer interaction. However, this tends to give naïve users an unrealistic conception of the intelligence of existing computer agents. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal sentiment analysis, wherein AI classifies the effects displayed by a videotaped subject.

Local search uses mathematical optimisation to find a solution to a problem. It begins with some form of guess and refines it incrementally. Gradient descent is a type of local search that optimises a set of numerical parameters by incrementally adjusting them to minimise a loss function. Variants of gradient descent are commonly used to train neural networks, through the backpropagation algorithm. Another type of local search is evolutionary computation, which aims to iteratively improve a set of candidate solutions by "mutating" and "recombining" them, selecting only the fittest to survive each generation. Distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimisation (inspired by bird flocking) and ant colony optimisation (inspired by ant trails).

Many problems in AI (including reasoning, planning, learning, perception, and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of tools to solve these problems using methods from probability theory and economics. Precise mathematical tools have been developed that analyse how an agent can make choices and plan, using decision theory, decision analysis, and information value theory. These tools include models such as Markov decision processes, dynamic decision networks, game theory, and mechanism design. Bayesian networks are a tool that can be used for reasoning (using the Bayesian inference algorithm), learning (using the expectation–maximisation algorithm), planning (using decision networks) and perception (using dynamic Bayesian networks). Probabilistic algorithms can also be used for filtering, prediction, smoothing, and finding explanations for streams of data, thus helping perception systems analyse processes that occur over time (e.g., hidden Markov models or Kalman filters).