貓咪智能項圈Cat Smart Collar

想法摘要Summary of the Idea

這份筆記記錄我對貓咪智能項圈的初步構想，以及目前查到的技術與成本資料。These notes record my early thinking on a cat smart collar, along with the technical and cost information I have found so far.

我目前想像的是一套居家系統，由輕量的感測項圈搭配家中的固定鏡頭，每天自動整理出貓咪的精華影片與活動報告。第一步傾向先不碰醫療與生理監測，盡量採購現有硬體，把力氣放在整合與軟體體驗。What I currently picture is a home system: a lightweight sensor collar paired with fixed cameras in the home, which together produce a daily set of highlight clips and an activity report for each cat. For a first step I lean toward leaving medical and physiological monitoring aside, buying existing hardware where possible, and putting the effort into integration and the software experience.

目前想到的方向大致有兩個。一是依感測資料自動剪輯影片並套用 AI 配音，降低飼主分享內容的門檻。二是把感測器與全景鏡頭累積的資料，轉成飼主平常看不到的貓咪生活資訊。空間區域管理只是其中一個例子，同一套資料能帶出的情境應該還有很多，下面先列幾個。Two directions stand out so far. The first is automatic clip editing driven by sensor data, with AI voiceovers that make sharing easy. The second is turning what the sensors and panoramic camera collect into information about the cat's life that owners normally never see. Zone management is only one example; the same data could support many more scenarios, a few of which are listed below.

一開始就可能抓住目光的情境Scenarios That Could Hook Owners Early

• 等門計時。Waiting at the door. 你回家前，牠在門口等了多久。情緒強、很適合分享。How long the cat waited by the door before you came home. Emotional and very shareable.
• 凌晨暴衝紀錄。Night zoomies. 半夜三點的家中跑酷，自動剪成片段。The 3 a.m. sprint around the house, cut into a clip automatically.
• 抓元凶。Who did it. 東西掉地上時，直接告訴你是哪一隻貓。When something hits the floor, the app tells you which cat did it.
• 多貓關係圖。Multi-cat relationship map. 誰最常黏誰、誰常追誰、誰在躲誰。Who sticks to whom, who chases whom, and who is avoiding whom.
• 牠的一天。A day in the cat's life. 睡在哪裡、睡了多久、一天動了多少。Where it slept, for how long, and how active it was.
• 窗邊與陽台停留。Windows and balconies. 牠在開著的窗邊或陽台待了多久、幾次。How long and how often the cat sat by an open window or on the balcony.
• 偷跳餐桌與流理台。Sneaking onto tables and counters. 你不在時牠上去了幾次。How many times it climbed up while you were out.
• 喝水與活動量變化。Changes in drinking and activity. 和上週比，喝水、活動是否明顯變少。Whether drinking or activity has dropped compared with last week.

底層構思：先讓問題被看見The Underlying Idea: Make the Problem Visible First

很多飼主平常覺得貓待在家裡很安全。就算裝了攝影機，每次拿起手機看的那幾秒剛好撞見危險畫面的機率其實很低，所以問題一直存在，只是沒被看見。Many owners assume their cat is perfectly safe at home. Even with a camera installed, the odds of happening to catch a dangerous moment in the few seconds they check their phone are very low. The problems exist; they just go unseen.

我的底層構思是創造需求。系統把不安或危險的情境自動抓出來、放到飼主面前，先讓飼主看見問題，再由我們提出付費的解答。這樣下一批產品，也就是「如何避免這種情況」，就有足夠的購買動力。My underlying idea is to create demand. The system catches uncomfortable or dangerous moments and puts them in front of owners, so they see the problem first, and then we offer a paid answer. That gives our next wave of products, the ones about preventing these situations, real purchase motivation.

可能的「問題到解答」對應，先作為發想：Some possible problem-to-solution pairings, as early ideas:

• 常在開窗或陽台停留，對應窗戶防護網或陽台封網Lingering at open windows or on the balcony, paired with window guards or balcony netting
• 常跳上爐台或流理台，對應爐台保護蓋或驅離裝置Jumping onto the stove or counter, paired with stove covers or deterrent devices
• 喝水次數偏低，對應智慧飲水機Low drinking frequency, paired with a smart water fountain
• 獨處時活動量偏低，對應自動逗貓器Low activity when alone, paired with an automatic cat toy
• 多貓搶食或追打，對應分開餵食與空間規劃產品Food stealing or chasing between cats, paired with separate feeders and space-planning products

有一點我覺得要先想清楚：這個模式成立的前提是推出來的都是真實發生的事。如果誤報太多或說得太誇張，飼主很快會覺得被推銷，反而傷到品牌信任，所以偵測準確度在這個模式裡很關鍵。One thing to settle early: the model only works if what we surface really happened. Too many false alarms or overdramatised alerts would make owners feel sold to and damage trust in the brand, so detection accuracy is central to this approach.

市場觀察Market Observations

台灣的貓市場看起來正在擴大。農業部 114 年調查推估全台家貓約 174 萬隻，兩年內成長 32.8%，首度超過家犬的約 146 萬隻。Taiwan's cat market appears to be growing. A 2025 survey by the Ministry of Agriculture estimates about 1.74 million pet cats, up 32.8% in two years and, for the first time, more than the roughly 1.46 million pet dogs.

同一份調查指出，成長主要來自多貓家庭增加。養狗家庭多半只養一隻，養貓家庭則常同時養兩隻以上，飼主以單身族與住在都會集合住宅的族群為多。The same survey attributes most of the growth to multi-cat households. Dog-owning homes usually keep one dog, while cat-owning homes often keep two or more, and owners skew toward single people living in urban apartments.

如果這個趨勢持續，我覺得有兩個可能的意義。第一，一個家庭可能會買多個項圈，客單價有機會提高。第二，多貓互動既是很受歡迎的影片題材，也可能是飼主想掌握的照護資訊。If the trend holds, I see two possible implications. One household may buy several collars, which could raise order value. And interaction between cats is both popular video material and, potentially, care information owners want.

產品定位的想法Thoughts on Positioning

目前想到的一句話主張是：The one-line pitch I have in mind for now:

每天自動收到你家貓今天的精華。Every day, automatically receive your cat's highlights.

我傾向把價值分成兩種來想。照護價值或許負責留住用戶：飼主不在家也能大致掌握飲水次數、是否進入危險區域、是誰打破了東西、兩隻貓相處是否融洽。創作價值或許負責帶進新用戶：系統自動剪出片段並套上 AI 貓咪心聲，分享出去的影片帶有品牌浮水印。I tend to think of the value in two parts. Care value may be what keeps users: even when away, owners get a rough picture of how often the cat drinks, whether it entered a dangerous area, who broke something, and whether two cats get along. Creative value may be what brings new users: the system cuts clips and adds an AI cat voiceover, and every shared clip carries a brand watermark.

我目前不太想主打「人人都能成為貓咪創作者」，因為多數飼主未必會天天發文。比較穩的說法或許是「飼主自己每天想看的內容」，分享擴散則是附帶效果。For now I would rather not lead with "everyone can become a cat creator", since most owners probably won't post every day. A safer framing may be "content owners want to watch for themselves", with sharing as a side effect.

系統架構構想Architecture Concept

目前的分工想法是：項圈只負責感測，拍攝交給家中插電的固定鏡頭。The current division of labour: the collar only senses, and filming is left to plugged-in cameras fixed in the home.

可能的組成部分Possible Components

1. 感測項圈。Sensor collar. 內含加速度計、氣壓計與藍牙低功耗晶片，重量希望控制在 10 公克左右，負責偵測動作、高度變化、與其他項圈的距離，並標示是哪一隻貓。An accelerometer, barometer and Bluetooth Low Energy chip, ideally around 10 grams, detecting motion, height changes and distance to other collars, and identifying the cat.
2. 家用鏡頭。Home camera. 一至兩台可水平轉動的鏡頭，插電持續循環錄影。One or two pan and tilt cameras on mains power, recording continuously in a loop.
3. 接收與雲端。Receiver and cloud. 接收項圈訊號，把事件時間點對到鏡頭錄影上，剪出前後片段並分類。Receives collar signals, aligns event timestamps with the camera footage, and cuts and labels clips.
4. 手機 App。Mobile app. 呈現精華片段與每日報告，提供配音、區域設定與提醒。Shows highlights and daily reports, with voiceover, zone settings and alerts.

或許可以再提供選配的項圈攝影機作為「冒險模式」，讓飼主偶爾拍貓咪的第一人稱視角，但不當主力。An optional collar camera could be offered as an "adventure mode" accessory for occasional point-of-view footage, but not as the main product.

為何傾向不把攝影機放在項圈上Why Not a Collar Camera

我一開始的構想是項圈直接長時間錄影，但查了資料後覺得有四個障礙。My first idea was a collar that records for long stretches, but the research points to four obstacles.

• 續航。Battery. 實測文章顯示，市面項圈攝影機連續錄影大多只有 60 到 120 分鐘。Hands-on tests show most collar cameras record for only 60 to 120 minutes.
• 重量。Weight. 同一份實測中最輕的專用款約 24 公克，而 Catlog 感測項圈只有約 9 公克。The lightest dedicated model in those tests weighs about 24 grams, while the Catlog sensor collar is about 9 grams.
• 省不了電。Power logic. 要剪到事件「之前」的畫面，鏡頭必須一直開著，感測器只能決定剪哪段。To capture footage from before an event, the camera has to stay on; sensors can only decide which part to keep.
• 畫面。Footage quality. 下巴視角容易拍到地板，而且衝刺跳躍時晃動最大。A chin-level view tends to show the floor, and shake is worst during the sprints and jumps we most want.

改由家用鏡頭拍攝後，依感測資料分類片段的概念還在。項圈資料可能正好補上鏡頭不擅長的部分，也就是「哪一隻貓」和「精確時間點」。With home cameras doing the filming, the idea of sorting clips by sensor data still holds. Collar data may fill exactly the gaps cameras struggle with: which cat it is, and precisely when something happened.

功能構想一：自動精華影片Feature Idea 1: Automatic Highlights

想法是依項圈訊號，從鏡頭錄影中自動剪出片段，大致歸入以下幾類。各類能不能實現，放在後面的技術可行性章節討論。The idea is to cut clips from camera footage based on collar signals, roughly sorted into the categories below. Whether each one is achievable is discussed in the feasibility sections later.

飼主端的流程我想像是三個動作：打開 App 看已分類的精華，選一段 AI 自動生成的貓咪心聲或自己輸入台詞，再一鍵分享到 TikTok、Instagram 或 YouTube Shorts。各平台的自動發文功能有審核或帳號類型限制，所以初期可能較適合用手機內建分享，而不是承諾全自動上傳。On the owner's side I picture three steps: open the app to see sorted highlights, pick an AI-generated cat voiceover or type a custom line, then share in one tap to TikTok, Instagram or YouTube Shorts. Auto-posting on those platforms comes with review or account-type limits, so at first the phone's built-in share sheet may be a better fit than promising fully automatic upload.

功能構想二：空間區域管理Feature Idea 2: Zone Management

想法是讓飼主自己把家中空間劃分成幾種區域，系統每天統計每隻貓在各區的停留時間，危險區則依飼主設定提醒。這可能是照護價值中最容易被飼主每天感受到的部分。The idea is to let owners divide their home into zones. The system would total each cat's time in every zone daily, and alert on dangerous zones according to the owner's settings. This may be the most tangible part of the care value day to day.

三種區域Three Zone Types

飼主怎麼劃定區域How Owners Would Define Zones

我希望一般飼主幾分鐘內就能設定完成，目前想到兩種互補的方式：鏡頭看得到的地方，在 App 的鏡頭畫面上用手指圈選並命名；鏡頭看不到的地方，貼一個藍牙小標籤。兩種方式的技術細節與成本，放在技術可行性三討論。I would like a typical owner to finish setup in a few minutes. Two complementary methods come to mind: for areas the camera can see, draw the zone with a finger on the live camera view in the app and name it; for areas it cannot see, stick a small Bluetooth tag there. Technical details and costs are in Feasibility 3.

每日區域報告Daily Zone Report

每隻貓每天一份報告，可能包括：One report per cat per day, possibly including:

• 在各偏好區的停留時間Time spent in each preferred zone
• 在飲水區的次數與總時間Number and total length of visits to the water area
• 進入限制區的次數與對應片段Entries into restricted zones, with clips
• 極危險區的事件紀錄Danger zone event log
• 與過去一週比較，變化明顯時提醒Comparison with the past week, with a nudge when something changes sharply

例如某隻貓這幾天到飲水區的次數明顯減少，飼主或許能提早發現。這也可能是「不喝水」這類照護需求在不碰醫療的前提下，最早能提供的基礎數據。If a cat's visits to the water area drop noticeably over a few days, for example, the owner might notice early. This may be the earliest basic data we can offer for care concerns like drinking, without touching medical features.

技術可行性一：動作與高度偵測Feasibility 1: Motion and Height Detection

初步看來，用項圈感測器偵測貓的跑跳已有不少學術驗證，但實驗室數字很高，推到不同貓、不同家庭時會下降多少仍是未知數。每種偵測我都先列兩到三條路徑。Research suggests collar sensors can already detect running and jumping in cats, but the high lab figures may drop when applied to different cats and homes, and by how much is still unknown. For each detection I list two or three possible paths.

追逐與高速移動Chasing and Fast Movement

路徑 A 的優點是鏡頭看不到也能偵測，路徑 B 只在畫面內有效，而且畫面中的速度會受距離影響。同一篇文獻回顧也提到，哪種模型最好仍是開放問題，不同研究的結論不一致。我的想法是項圈負責觸發，鏡頭負責確認。Path A works even outside the camera's view, while path B only works in frame and its speed reading depends on distance. The same review notes that the best model is still an open question, with studies reaching conflicting conclusions. My current thinking: the collar triggers, the camera confirms.

跳躍與爬上爬下Jumping and Climbing

以 ±25 公分的精度推算，跳上一公尺以上的貓跳台或冰箱應該偵測得到，跳上約 40 公分的沙發則可能處在邊緣。另外我推測開關門、冷氣或窗外風壓可能造成氣壓干擾，這點沒查到資料，需要實測。比較穩的做法或許是加速度計偵測到跳躍動作後，再看前後幾秒的氣壓差確認高度。At ±25 cm accuracy, a jump onto a cat tree or fridge above one metre should be detectable, while a jump onto a 40 cm sofa may sit at the margin. I also suspect doors, air conditioning and wind could disturb pressure readings; I found no data on this, so it needs testing. A steadier approach may be to let the accelerometer detect the jump, then confirm height from the pressure difference over the surrounding few seconds.

技術可行性二：兩貓互動與個體辨識Feasibility 2: Interaction and Identifying Each Cat

「兩個項圈靠近就算互動」這個想法，查完資料後我覺得只在距離很近時比較可靠。藍牙訊號強度換算距離的誤差不小，而且貓的身體會擋住訊號。After reading up, I think "two collars close together means interaction" is only reliable at very short range. Converting Bluetooth signal strength to distance carries sizeable error, and the cat's own body blocks the signal.

兩貓互動Interaction

我目前的想法是：第一步用 A 判斷「非常靠近」，再用 B 在鏡頭範圍內確認。C 精度最好，但成本與耗電都高，或許留到後續版本。My current thinking: start with A to detect "very close", then confirm with B inside the camera's view. C is the most accurate but costly and power-hungry, so it may belong in a later version.

是哪一隻貓Which Cat Is It

從商品來看，純影像分辨個體已經有人在做，但我們自己做的門檻不低。有項圈的話，路徑 A 或許就是我們相對純鏡頭產品的優勢。Commercial products show that vision-only identification is already being done, though building our own would not be trivial. With a collar, path A may be our edge over camera-only products.

技術可行性三：區域劃定與危險區警報Feasibility 3: Zones and Danger Alerts

區域判斷我目前看到三條路徑：純視覺的畫面判斷、以項圈作為實體定位的藍牙偵測，以及精度最高但最貴的 UWB。各自的強項不同，或許需要互相搭配。I see three paths for zone detection: vision-only judgement from the camera image, Bluetooth detection using the collar as a physical locator, and UWB, which is the most precise and the most expensive. Each has different strengths and they may need to be combined.

路徑 A：純視覺Path A: Vision Only

做法。How. 飼主在鏡頭畫面上圈出區域，鏡頭辨識出貓後，判斷貓的位置是否落在圈選範圍內，並計算停留秒數。The owner draws a zone on the camera view. Once the camera detects the cat, it checks whether the cat falls inside the zone and counts how long it stays.

依據。Evidence. 市售鏡頭已有類似介面，例如 Tapo C210 可自訂多個活動區域與隱私遮蔽區，但它的區域只搭配一般動態偵測，沒有專門的寵物辨識。算力方面，RV1106 等級晶片跑小型 YOLO 約每秒 10 至 15 張，對「停留幾秒」的判斷應該夠用。Retail cameras already offer similar setup: the Tapo C210 lets users draw several activity zones and privacy block zones, though its zones only work with generic motion detection, not pet recognition. On compute, a small YOLO model runs at roughly 10 to 15 frames per second on an RV1106-class chip, likely enough to judge a dwell of a few seconds.

成本。Cost. 不需額外硬體，成本在鏡頭本身（見成本表）。No extra hardware; the cost sits in the camera itself (see the cost table).

限制。Limits. 只在畫面內有效，陽台、其他房間等看不到的地方無法判斷。影像是平面的，貓站在爐台前的地上與站在爐台上，可能在畫面中重疊，這是我的推論，需要實測。Only works within view; a balcony or another room cannot be judged. The image is flat, so a cat on the floor in front of the stove and a cat on the stove may overlap on screen. That is my own inference and needs testing.

路徑 B：項圈藍牙實體定位Path B: Collar as a Bluetooth Locator

這條路徑有兩種實作方式。This path has two possible implementations.

B1 區域標籤。B1 Zone tags. 在區域貼一個鈕扣電池的藍牙標籤，項圈掃描到訊號夠強就判定在區域內。市售例子 MokoSmart M1 尺寸約 26 × 26 × 7 公釐、重 5.4 公克，用 CR2032 預設可用約 1.5 年。以公開的耗電參數粗算，nRF52832 每秒廣播一次平均約 9.5 微安，200 毫安時電池約可用 2.4 年。A coin-cell Bluetooth tag is placed at the zone; when the collar picks up a strong enough signal, the cat counts as inside. A retail example, the MokoSmart M1, measures about 26 by 26 by 7 mm, weighs 5.4 g and runs about 1.5 years on a CR2032 by default. A rough calculation from published power figures: an nRF52832 advertising once per second averages about 9.5 µA, giving roughly 2.4 years on a 200 mAh cell.

B2 房間接收器。B2 Room receivers. 反過來讓項圈只負責廣播，每個房間插一個小型接收器，訊號最強的接收器就代表貓所在的房間。開源專案 Bermuda 就用這個方式，以 ESP32-C3 做接收器，每顆約 2.5 至 3 美元，三房住宅總硬體約 20 美元以內，預設每秒更新。項圈只廣播不掃描，理論上比 B1 省電。Alternatively, the collar only advertises and a small receiver plugs into each room; the receiver hearing the strongest signal marks the room. The open-source Bermuda project works this way using ESP32-C3 receivers at about US$2.50 to 3 each, under US$20 of hardware for a three-bedroom home, updating every second by default. Since the collar advertises without scanning, it should in theory use less power than B1.

依據。Evidence. 藍牙定位大約是房間級，Bermuda 作者自己說明不是公尺以下的定位；研究也顯示牆壁會造成約 ±2 公尺誤差，但距離 3 公尺以內的配置可改善到 8 至 10 公分。Catlog 本身也是「項圈搭配基地台」架構，一台基地台約涵蓋 60 平方公尺、最多四個項圈，但評測提到資料常有缺漏。Bluetooth positioning is roughly room-level; Bermuda's own documentation says it is not sub-metre. Research shows walls can cause about ±2 m error, improving to 8 to 10 cm when distances are kept within 3 m. Catlog itself uses a collar plus base station design, covering about 60 square metres and up to four collars per station, although a review notes frequent gaps in the data.

限制。Limits. 適合判斷「在不在陽台」「在哪個房間」這種較大的範圍，不太適合區分「在爐台上」還是「在爐台旁」。Suited to coarse questions like "is the cat on the balcony" or "which room", less so to telling "on the stove" from "next to the stove".

路徑 C：UWB 定位Path C: UWB Positioning

做法。How. 房間內放置至少三個 UWB 基地台，項圈內放 UWB 晶片，算出平面座標。Place at least three UWB anchors in a room and a UWB chip in the collar to compute a 2D position.

依據。Evidence. Qorvo DWM3001C 模組標示平面精度優於 15 公分，電池壽命在雙向測距模式下以月計。新一代低功耗 UWB 宣稱可用鈕扣電池數年，但仍在發展中。Qorvo's DWM3001C module lists 2D precision under 15 cm, with battery life measured in months in two-way ranging mode. Newer low-energy UWB claims multi-year coin-cell life but is still emerging.

成本。Cost. 項圈端一顆模組約 17 至 24 美元，再加上每個房間三個以上的基地台，整體成本明顯高於 A 與 B。About US$17 to 24 per module on the collar, plus three or more anchors per room, clearly more expensive than A or B.

目前的想法Current Thinking

或許可以是：鏡頭看得到的區域用 A，看不到的區域與房間層級用 B，兩者都涵蓋的危險區用兩種訊號交叉確認以減少誤報。C 或許留給高階版本。對飼主溝通時，精度最好說成「區域級」，避免讓人以為是公分級。One possibility: use A for zones the camera can see, B for out-of-view zones and room level, and cross-check both signals for danger zones covered by both, to cut false alarms. C could be kept for a premium version. When talking to owners, it may be best to describe precision as zone-level, so nobody expects centimetre accuracy.

技術可行性四：拍攝、剪輯與配音Feasibility 4: Filming, Clipping and Voiceover

鏡頭硬體Camera Hardware

路徑 A 最快，但很多成品的影像封閉在自家雲端，要先確認能不能取得畫面。路徑 B 比較自主，也能避開中國雲端的資安疑慮，但需要韌體工程能力。Path A is fastest, but many retail cameras lock video inside their own cloud, so access has to be confirmed first. Path B gives more control and avoids concerns over Chinese cloud services, but needs firmware engineering.

鏡頭追蹤Camera Tracking

• A 內建動態追蹤。A Built-in motion tracking. 市售鏡頭多半已有，但追的是「任何在動的東西」，不一定是貓。Most retail cameras have it, but it follows anything moving, not necessarily the cat.
• B 事件驅動轉向。B Event-driven presets. 飼主設定區域時順便記住鏡頭角度，項圈或區域觸發事件時就轉到該預設位置。這是我的構想，尚未查到現成產品。When owners set up a zone, the camera saves that angle as a preset and turns to it when a collar or zone event fires. This is my own idea and I have not found an existing product doing it.

片段擷取Clip Extraction

• A 本地循環錄影。A Local loop recording. 鏡頭持續錄到記憶卡，事件發生時只上傳前後幾秒。Tapo C210 支援最大 512 GB 記憶卡。節省頻寬與雲端空間。The camera records continuously to a memory card and uploads only a few seconds around each event. The Tapo C210 supports cards up to 512 GB. This saves bandwidth and cloud storage.
• B 全部上雲。B Upload everything. 影像全部上傳雲端再剪，邏輯簡單，但頻寬與儲存成本高，實際金額還沒估。Upload all video and cut in the cloud. Simpler logic, but higher bandwidth and storage cost, not yet estimated.

時間對齊也要考慮：藍牙接收器預設約每秒更新一次，所以片段前後應該留幾秒緩衝。Timing also matters: Bluetooth receivers update about once per second by default, so clips should probably include a few seconds of padding on each side.

AI 配音AI Voiceover

一句約 50 字的貓咪心聲，按上面推算成本大約在 0.0002 到 0.002 美元之間，配音本身應該不是成本重點。各家的中文語音質感、以及產生台詞的語言模型成本，我還沒查。At these prices, a 50-character cat line costs roughly US$0.0002 to 0.002, so voiceover itself is unlikely to be a major cost. I have not yet checked Chinese voice quality across providers, nor the language model cost of writing the lines.

元件單件成本粗估Rough Unit Cost of Components

以下是目前查到的單件價格，以美元計。電子元件多半取自電子零件通路商 LCSC 的網路報價，實際代工報價可能不同。印刷電路板、充電電路、外殼、項圈帶與組裝工費都還沒查到可靠數字，需要向代工廠詢價。These are unit prices found so far, in US dollars. Most electronic parts come from online quotes at the distributor LCSC, and actual factory quotes may differ. Circuit boards, charging circuitry, housings, collar bands and assembly labour have no reliable figures yet and need factory quotes.

項圈端Collar

加總起來，不含 UWB 的項圈主要元件大約在 5.5 到 12 美元之間，還要加上電路板、外殼與組裝。Adding these up, the main collar components without UWB come to roughly US$5.50 to 12, before boards, housing and assembly.

居家端Home Side

服務端Service Side

市售鏡頭的零售價已含廠商利潤，實際代工成本應該更低，但尚未查證。Retail camera prices already include the brand's margin, so actual factory cost is probably lower, but this is unverified.

競爭格局Competitive Landscape

從目前看到的產品來看，對手各自做一塊，我還沒看到把感測、影像、剪輯與區域管理放在一起的。不過我的搜尋範圍有限，可能有遺漏。From the products I have seen, each competitor covers one piece, and I have not yet found one combining sensing, video, editing and zone management. My search was limited, though, so I may have missed some.

如果硬體向中國採購，別人也可以做出類似的東西，所以我目前覺得優勢或許不在硬體，而在整合體驗、雙方在台灣寵物圈的通路與 KOL 人脈，以及分享影片帶動新用戶的內容飛輪。If hardware is sourced from China, others can build something similar, so for now I think our edge may lie not in hardware but in the integrated experience, our network and KOL relationships in Taiwan's pet industry, and a content flywheel where shared clips bring in new users.

商業模式的想法Thoughts on the Business Model

我目前傾向「硬體帶進用戶，訂閱帶來獲利」。硬體容易被複製、毛利也薄，穩定收入或許應該來自訂閱。參考價格：Catlog 的訂閱在亞馬遜標示每月 8 美元起，評測文章則寫每隻貓每月 14 美元；Furbo 的訂閱年繳後每月約 6.99 美元起。For now I lean toward hardware for acquisition and subscriptions for profit. Hardware is easy to copy and thin on margin, so steady income may need to come from subscriptions. For reference, Catlog's subscription is listed from US$8 a month on Amazon, while a review cites US$14 per cat per month; Furbo's subscription starts at about US$6.99 a month when billed yearly.

硬體可能的品項Possible Hardware Items

• 入門組合：一個項圈搭配一台鏡頭Starter kit: one collar and one camera
• 加購項圈：多貓家庭的主要加購Extra collars, the main add-on for multi-cat homes
• 區域標籤或房間接收器Zone tags or room receivers
• 可更換的項圈帶與品牌聯名款，參考 Swatch 換錶帶的做法，未來或許能與 Bugaboo 等品牌合作Swappable bands and co-branded editions, following Swatch's strap model, perhaps with brands such as Bugaboo later
• 選配項圈攝影機Optional collar camera

訂閱可能的切分Possible Subscription Split

免費版或許提供當日精華、基礎區域統計與極危險區警報。危險警報關乎安全，我傾向不收費，以免影響品牌形象。付費版則可以是影片長期保存、AI 配音額度、完整每日報告與異常提醒。定價都還沒有想法。A free tier might include daily highlights, basic zone stats and danger zone alerts. Since danger alerts concern safety, I lean toward not charging for them, to protect the brand. A paid tier could cover long-term clip storage, AI voiceover credits, full daily reports and anomaly alerts. Pricing is still entirely open.

可能的風險Possible Risks

可以怎麼驗證How This Could Be Tested

我的想法是先用低成本的方式確認需求與技術，再決定要不要投入打樣。先後順序大致如下，每一步的結果都可能改變後面的方向。My thinking is to check demand and technology cheaply first, then decide whether to fund prototyping. The rough order is below, and each step's results could change the direction of what follows.

1. 需求驗證。Demand check. 採購現成寵物鏡頭，從雙方人脈找幾個多貓家庭試用，人工剪片、配音、手動做區域報告，再請 KOL 發布測試。Buy off-the-shelf pet cameras, recruit a few multi-cat households through our networks, cut clips, add voiceovers and build zone reports by hand, then have KOLs post test clips.
2. 感測器驗證。Sensor check. 用開發板搭配加速度計與氣壓計，讓幾隻貓配戴，對照鏡頭畫面標註，實測跳躍與高度判斷的準確度，也確認氣壓干擾。Fit a few cats with dev boards carrying an accelerometer and barometer, label against camera footage, and measure jump and height accuracy, including pressure interference.
3. 區域驗證。Zone check. 同一個家中同時試純視覺與藍牙兩條路徑，比較誤報與漏報。Run vision-only and Bluetooth paths side by side in the same home and compare false and missed alerts.
4. 鏡頭整合驗證。Camera integration check. 確認候選鏡頭是否能取得影像、能否依事件轉向與剪片。Confirm candidate cameras expose video and can turn and clip on events.
5. 代工詢價。Factory quotes. 帶著上面的結果向深圳代工廠詢價，補齊電路板、外殼與組裝成本。Take these results to Shenzhen factories for quotes covering boards, housing and assembly.

需求驗證時，我想觀察的是：每天能剪出多少可用片段、貓的配戴接受度、飼主是否每天看報告、KOL 片段的互動與一般貓片相比如何，以及付費意願。During the demand check I would watch how many usable clips come out per day, how well cats tolerate the collar, whether owners open the report daily, how KOL clips perform compared with ordinary cat videos, and willingness to pay.