投稿時間:2022-03-11 22:21:44 RSSフィード2022-03-11 22:00 分まとめ(22件)

カテゴリー等	サイト名等	記事タイトル・トレンドワード等	リンクURL	頻出ワード・要約等/検索ボリューム	登録日
IT	ITmedia 総合記事一覧	[ITmedia News] ポケモンGOで13日に「サンド」と「サンド（アローラのすがた）」大量発生　色違いも2種類	https://www.itmedia.co.jp/news/articles/2203/11/news185.html	itmedia	2022-03-11 21:45:00
python	Pythonタグが付けられた新着投稿 - Qiita	discord.pyでボイスチャンネルにいるメンバーを取得する	https://qiita.com/NightWatchWife/items/569d9a5d4cb7d094d1f3	discordpyでボイスチャンネルにいるメンバーを取得するボイスチャンネルにいるメンバーを取得する方法について色々とたらい回しにされたので参考までに書いておきます。	2022-03-11 21:55:33
python	Pythonタグが付けられた新着投稿 - Qiita	RaspberyPi Zero WHとSunlightセンサで紫外線などを測定してみた	https://qiita.com/kosukein38/items/bad23d96386d5698675f	準備したもの以前の記事にも記載していましたが、Sunlightセンサを追加して、再掲します。	2022-03-11 21:19:52
python	Pythonタグが付けられた新着投稿 - Qiita	YahooNewsが毎朝届くLINEBOT	https://qiita.com/motty00/items/cb07f9f3042454c576a9	pythonの勉強もヶ月程なので冗長なコードがありますが、暖かく見守ってください笑作成の経緯定期的にLINEがお知らせをしてくれるBOTを作りたいなと思い立った。	2022-03-11 21:06:57
Ruby	Rubyタグが付けられた新着投稿 - Qiita	[Devise] 更新をpasswordなしで可能にする	https://qiita.com/kandalog/items/cd39599b75e3368e7fb7	Devise更新をpasswordなしで可能にするはじめに認証機能で有名なRubyonRailsのgemDeviseを使ったので記事を残そうと思います。	2022-03-11 21:21:38
AWS	AWSタグが付けられた新着投稿 - Qiita	Elastic IP	https://qiita.com/daisan182/items/8826b667b88b57026459	補足パブリックIPとの違いECインスタンスが停止した場合、IPアドレスが変更される。	2022-03-11 21:58:54
AWS	AWSタグが付けられた新着投稿 - Qiita	AmplifyのIAMでauth-roleとunauth-roleってなに?	https://qiita.com/okapee0608/items/52b240044f90d8497dda	authRoleunauthRoleとは何者なのかauthRoleとunauthRoleは一番最初にamplifyinitをしたときに生成されるIAMのポリシーだそうです。	2022-03-11 21:22:48
技術ブログ	Mercari Engineering Blog	Elasticsearch運用ノウハウ	https://engineering.mercari.com/blog/entry/20220311-97aec2a2f8/	elasthellip	2022-03-11 13:42:13
海外TECH	DEV Community	How to Implement AI Self Checkout in Retail if You Are Not Amazon	https://dev.to/liubovzatolokina2022/how-to-implement-ai-self-checkout-in-retail-if-you-are-not-amazon-3ki7	How to Implement AI Self Checkout in Retail if You Are Not AmazonOnline retail has one key advantage ーcustomer experience No queues no delays and little movement to make a purchase According to a research from Forrester of U S retail sales will still occur in bricks and mortar stores because people want to interact with a product before buying or simply don t want to wait for delivery The idea of checkout free shopping in venues crystalized as Amazon Go Tesco Walmart and many more The idea of using fully automated checkout with computer vision is a successful example of retail automation But a few store owners want to build a whole new outlet to run their business offline As it requires an integrated software infrastructure as well as imposes development and financial challenges we will discuss today In this article we ll analyze how any brick and mortar store can be automated with computer vision systems Here we ll look at how it works what the options for checkout automation are and what challenges are out there Image credit Computer vision checkout automation for brick and mortar retailThe majority of in store operations like shelf management checkout or product weighing require human supervision Human productivity is basically a performance marker for the retailer and it often becomes a bottleneck as well as becoming a customer frustration factor Namely checkout queues are the pain point both for customers and retailers But it s not only the queues since actual human effort costs money So how does computer vision apply to these operations Computer vision CV is a technology under the hood of artificial intelligence that enables machines to extract meaningful information from the image At its core computer vision aims at mimicking human sight So analogically to an eye CV relies on camera sensors that capture the environment In its turn an underlying neural network it s brain will recognize objects their position in the frame or some other specific properties such as differing a Pepsi can from Dr Pepper can That s our ground base for understanding how computer vision can fit brick and mortar retail tasks as it can recognize products situated in the frame These products can be placed on the shelves or carried by the customers Which allows us to exclude barcode scanning cash register operation or self checkout machines Although implementations of computer vision significantly differ by complexity and budgeting there are two common scenarios of how it can be used for retail automation So first let s look at how full store automation can be built AI powered autonomous checkout full store automationAutonomous checkout is called by different names “cashierless “grab and go “checkout free etc In the shopping experience of Amazon Tesco and even Walmart such stores check the products during the shopping and charge for them when you walk out Sounds simple and that s how it works in a basic scenario Shopping session start Shops like Amazon use turnstiles to initiate shopping via scanning a QR code At this point the system matches the Amazon profile and digital wallet with the actual person entering the store Person detection This is the recognition and tracking of people and objects done via computer vision cameras Simply cameras remember who the person is and once they take a product from the shelf the system places it into a virtual shopping cart Some shops use hundreds of cameras to view from different angles and cover all the store zones Product recognition Once the person grabs something from the shelf and takes it with them cameras capture this action Matching the product image on video with the actual product in the retailer s database the store places an item into a virtual shopping cart Checkout As the product list is finished the person may just walk out When the person leaves a zone covered by cameras computer vision considers this as the end of a shopping session This triggers the system to calculate the total sum and charge it from the customer s digital wallet From the customer standpoint such a system represents a similar shopping experience as it is in the online stores except you don t need to checkout Enter find what you want grab it and leave Although to provide customers with full autonomy and cover all the edge cases we ll need to solve a large number of problems technically So what s so complex about autonomous checkout The challenges of AI powered autonomous storesCustomer behavior can be unpredictable as we are going to automate checkout for dozens of people that check and buy thousands of products at the same time This imposes a number of challenges for computer vision CONTINUOUS PERSON TRACKINGAs the customer enters the store the system should be able to continuously track them along shopping routes We need to know that it s the same person who took this or that item in different parts of the store In a crowded store continuous tracking might be difficult As long as it s not allowed to use face recognition the model should recognize people by their appearance So what will happen if somebody takes off his coat or carries a child on shoulders To enable continuous tracking we ll need to provide coverage for cameras to detect people passing from zone to zone Placing cameras at different angles we also need sensors to communicate their precise location so we can use this data to track objects more accurately THE “WHO TOOK WHAT PROBLEMThen we have to remember there are also products right And customers shopping process is not linear They move items smell them put them back and go to another shelf Especially when there are multiple people at one shelf it becomes difficult for a model to recognize who took what and if they actually took the product to buy Amazon for example solved this problem by implementing human pose estimation and human activity analysis Basically that s another layer of artificial intelligence coupled with computer vision What it does is it measures the position and movement of a person to predict what he or she grabs and if the product was taken to be purchased This solves the problem with multiple customers at a shelf and helps to denote who took this specific product even if the camera was blocked by somebody IDENTIFYING SIMILAR PRODUCTSConcerning products we ll also need to deal with similar packages Some products have minor differences in their look which makes it harder for the model to fetch all the detail Especially if there is some obstruction going on in the frame or the object is moving fast We can address this issue through training the model to spot little details and use cameras with higher resolution and frame rate While it looks beneficial to use autonomous checkout the complexity of such a system can be onerous For a tech first company this is not a problem But for the usual retailer the burden brought by artificial intelligence lowers the value of such automation That s why partial store automation with computer vision can be more suitable Smart vending machines partial store automationWhen it comes to vending machines they can be placed in store or moved out to other indoor and outdoor locations And this can be an elegant solution to the problem imposed by tracking the whole store Vending machines can be represented by shelves with glass doors or regular fridges using computer vision cameras to operate purchase processes Installing a QR code scanner we can minimize the checkout procedure to the location of a single fridge So the idea is quite simple Shopping session start The session starts once a person approaches the fridge and opens it up This can be done via scanning a QR via mobile app if it s a door closed fridge In the case of a usual shelf cameras can track what s grabbed from it to initiate the session Creating a virtual shopping cart As the person scans the QR code it s a signal for a system to create a shopping cart for this specific user Product recognition The cameras might be installed inside or outside of the vending machine The internal cameras should be able to track the taken put back products External cameras might track manipulations within an open fridge just like with a regular shelf Both types of cameras capture the products and put them into a shopping cart As the person might examine multiple items and move from side to side CV cameras can also track the person in the frame This will help us verify that it s a single person making a purchase and not another one standing nearby Verifying products When the product is taken the system sends this data to compare the image of the product with the one in the database and extract the price Additionally we can update availability automatically in our inventory management system Editing product list Once the products are taken they will be sent to the user s shopping cart available on their smartphone or tablet on the fridge Here the customer can modify items and proceed to the payment Checkout In case of a mobile application and QR code scanning closing the fridge might be a trigger point to complete a purchase and charge a sum from a digital wallet But there might also be a POS terminal installed to allow credit card payment At this point the purchase is done and the person can leave the store While it looks like a relatively weak alternative to the autonomous checkout system vending machines can be scaled easily to automate the whole store Which makes a little difference in terms of customer experience but requires less engineering effort and budgeting The same concept of modular automation can be applied to numerous other cases Except for supermarkets and grocery stores computer vision kiosks can also be installed in food service venues or coffee shops CHECKOUT FREE FOOD SERVICERestaurants cafes and canteens often use a buffet serving system like a sideboard with portioned dishes customers can choose from Customers place dishes on trays then need to check out their order which can potentially be handled by a computer vision kiosk A machine learning model sitting on the backend can be trained to recognize dishes and other products placed on the tray to launch the checkout process This idea can be implemented as a checkout kiosk where a set of cameras will scan the order The actual payment can be completed via a usual POS terminal or using a mobile application and a digital wallet The concept of cashierless operations can be taken to extremes like with Starbucks Using Amazon s system Starbucks became the first of a kind grab amp go coffee shop Customers can place an order via a mobile application and come for their coffee without any checkout similar to Amazon GO However handling computer vision projects requires knowledge of a subject matter Specifically data science and machine learning expertise So now let s talk a bit of what you should know to approach computer vision based checkout automation How to approach AI based checkout Based on our experience let s examine the steps it takes to create a computer vision system for automation in retail We ll focus on the smart fridge case as the most approachable and versatile one GATHERING REQUIREMENTSFirst of all we need to understand our business case in detail Preferred automation method Choosing between smart fridges or other types of dispenser machines might require less global modifications to the store while maintaining a scalable approach Full store automation will mostly require changes to the venue layout and additional hardware like turnstiles which can be a con for the majority of the store owners Store size Vending machines can be installed in basically any number to cover all of the store s inventory and product diversity So the store size will determine how many vending machines you ll need and what will be the store layout using smart fridges for some part of products Quantity of products for recognition As any other machine learning project a computer vision system requires training before it can recognize anything A single fridge might contain to different products So we should consider those numbers as it will determine how long the training phase will take Existing infrastructure In most cases physical stores don t have enough integration between inventory management point of sale and accounting Although computer vision systems will require access to the store data to automate sales updates and product availability So examining your existing infrastructure is another point to understand when considering the requirements of this project So let s say a single fridge can contain items and we ll focus on those numbers DATA COLLECTIONComputer vision is an artificial intelligence technology Which means we need data so it can recognize objects The data is used for model training to identify different products in the frame as well as identify people and what they grab The optimal way to collect data for object recognition is basically to record each product on video from different angles and lightning conditions It is important to have these videos categorized by product so the labeling what product is in the frame will be done automatically General recommendations for gathering the data are that it should be as close as possible to how it will look for real users Once we implement a working model to automate checkout we ll need frames per second This is required to guarantee fast operation of the model The higher the frame rate the smoother the image is and the more detail we can extract from it MODEL TRAININGThe next step is training Once we collect all the video recordings a machine learning expert will prepare them for model training This process can be split into two tasks Preparing data means we need to split all the video frames into separate images and label the products we need to detect Put simply we extract photos out of a minute long video and draw bounding boxes around our target objects Choosing an algorithm An algorithm is a mathematical model that learns patterns from the given data to make predictions For tasks like object recognition there are existing working algorithms that can be applied for building a model So our task here is to choose a suitable one and feed it with our data The process of training may take several weeks as we struggle to get decent accuracy MODEL RETRAININGIf any products are added or swapped in the process the model needs to be retrained This is because prediction results will differ depending on the data input This means that each time a store obtains new items for sales and places them into a computer vision fridge ーwe ll need to launch a new training phase for the model to learn new items Given that we ll need retraining to recognize say Pringles cans on the image if there weren t any Pringles before Although this becomes easier as soon as we implement cameras in the fridge because we can use live recordings to make annotations and launch training again REQUIRED INFRASTRUCTUREThe existing infrastructure in the store is usually represented by a server that processes inventory updates and records sales volume via POS terminals To implement a machine learning model we ll need to add several components Cameras to record and pass the visual data Video processing unit This can be a video card or a single board computer like the Nvidia Jetson that includes a GPU optimized for computer vision needs QR scanner This sticker is placed on a turnstile or a fridge the user scans to identify the person and launch the shopping process Model server As we re talking about real time video processing implementing a hardware server at the store will guarantee more stable results Basically as a person grabs something from a fridge the reaction of the system should be noteless so that hardware components can respond fast enough All of those components should be interconnected as there has to be data flow between each unit As for the cameras we also want to make sure the store has a stable and fast bandwidth Since cameras will process live streams of data in the real time there has to be no delay for the model to function properly On the other hand the customer will expect a fast reaction of the vending machine which depends on how quickly the model receives and processes the data Daniil LiadovPython engineer at MobiDev PRIVACY CONCERNSAmong other questions that might concern both retailers and customers is privacy Since computer vision is designed to detect and track objects on video recording and storing such data may violate the privacy laws in some countries Although in the US it s generally legal to use surveillance cameras in stores As long as customers are tracked with random IDs just for the sake of the checkout task no other technologies like face recognition are required And even if the camera captures a person s face it could be blurred using AI to sustain confidentiality Is AI self checkout for every retailer All with all systems autonomous checkout may seem like a pricey and bulky thing to implement Customers are still willing to use more convenient checkout methods however That s noted in a Retail Customer Experience report that of consumers would choose self checkout over interaction with a cashier That being said vending machines might be an affordable option for the retail industry as it brings a lot of benefits for a reasonable cost Additionally such systems can be customized to serve the specific needs of a given retailer due to flexibility of machine learning models Basically any type of product can be recognized with proper training So convenience stores are not the only ones who can benefit from computer vision applications	2022-03-11 12:12:25
海外TECH	Engadget	The Morning After: Nintendo’s Super Mario theme park is coming to the US	https://www.engadget.com/the-morning-after-nintendos-super-mario-theme-park-is-coming-to-the-us-121522139.html?src=rss	The Morning After Nintendo s Super Mario theme park is coming to the USThe Nintendo theme park experience is headed to the US Universal Studios Hollywood announced Super Nintendo World ーa ride and interactive area in the style of the Super Mario game series ーwill debut at the California theme park in Nintendo s debut park in Osaka was delayed by nearly a year due to the pandemic opening to limited numbers in March For the US spin fans can expect an interactive area a special themed ride and themed shopping and dining ーbecause the park has to make money right Like the Osaka iteration Super Nintendo World will be an expansion of the current Universal Studios Hollywood marking the first major expansion of the Hollywood park since the Wizarding World of Harry Potter was added in If your heart is still set on seeing the Japanese original it will be getting a Donkey Kong expansion in ーMat SmithThe biggest stories you might have missedGoogle s latest Android Messages updates include iOS reactions and YouTube previewsThe best mobile microphones you can buy plus how to pick oneApple s K Studio Display should support Windows including the webcam and speakers Android update offers voice pay for parking and offline live transcriptions Valkyrie Elysium is an action RPG sequel to PS s Valkyrie Profile Call of Duty Warzone is coming to mobileActivision is currently hiring for new roles to build the game Call of Duty Warzone the free to play battle royale will soon have a mobile version In a tweet the game s publisher Activision announced it was hiring for a slate of new mobile roles It s not the first CoD nbsp title adapted for mobile ーActivision released the kinda OK Call of Duty Mobile in Continue reading Ford s Maverick pickup is perfect for nerdsLow cost D printing and customization EngadgetThe only problem is that EVs are pricey So why not embrace a hybrid that also happens to be a small truck with a very impressive starting price The Ford Maverick starts at and ships with a hybrid powertrain that delivers up to MPG In addition to being a truck it has a bed built for customization Plus thanks to makers like Robert Trapp the FITS Ford Integrated Tether System already has D designs for printing or manipulation It s relatively cheap customizable and a hybrid Roberto Baldwin gives it a test drive Continue reading NVIDIA s high end GeForce Now streaming tier is available on a monthly planIt costs per month or for six months There s finally a month to month payment option for GeForce Now RTX Before NVIDIA only offered six months of access for Now it costs per month to try that tier That lowers the barrier to entry though you ll save more in the long run with the six month plan Expect p resolution gaming with ray tracing at up to fps on Mac and PC and K HDR resolution at fps on NVIDIA Shield Continue reading Red Rocks Amphitheater will no longer use Amazon s palm scanning techActivists and artists pressured Denver Arts and Venues to stop using the system Red Rocks Amphitheater one of the most recognizable concert venues in the US no longer plans to use Amazon s palm scanning technology for ticketless entry Activists and artists including Fight for the Future Tom Morello Rage Against the Machine and Kathleen Hanna Bikini Kill pressured Denver Arts and Venues to refrain from using Amazon One at the venues it manages Those who signed an open letter cited concerns about Amazon sharing palmprint data with government agencies that seek to track marginalized people and activists Continue reading Samsung adds performance throttling controls to the Galaxy SThey re only available in South Korea for now Users on Samsung s Korean community forums are receiving an update for the Galaxy S series that adds a quot Game Performance Management Mode quot to Game Booster The release should let users override the throttling feature and wring more speed out of the flagship phones at least so long as they re willing to accept reduced battery life Continue reading DuckDuckGo reverses course will demote Russian propaganda in search resultsThe founder said he is “sickened by Russia s invasion of Ukraine The search engine DuckDuckGo will down rank sites that spread Russian propaganda and disinformation Founder and CEO Gabriel Weinberg tweeted that the privacy focused search engine would be releasing updates to ensure Russian disinformation sites rank further down in search results Earlier this month DuckDuckGo announced it would pause its relationship with Russian state owned search engine Yandex Continue reading Razer s Seiren lapel mic works over BluetoothIt s made for streamers on the move RazerRazer s Seiren Bluetooth lapel microphone packs an omnidirectional mic and AI based noise suppression The lapel mic includes a mm jack for monitoring through headphones and you can customize it through the Razer Streaming App to tweak the noise suppression level making for a pretty compelling mobile mic experience at least on paper The Seiren Bluetooth is available for and should work with quot all quot phones as well as common streaming apps like Streamlabs Twitch and YouTube Continue reading	2022-03-11 12:15:22
ニュース	BBC News - Home	Ukraine war: People can welcome refugees into their homes - PM	https://www.bbc.co.uk/news/uk-60701941?at_medium=RSS&at_campaign=KARANGA	boris	2022-03-11 12:28:00
ニュース	BBC News - Home	‘It was a nightmare in that place’	https://www.bbc.co.uk/news/uk-northern-ireland-60698025?at_medium=RSS&at_campaign=KARANGA	apology	2022-03-11 12:27:59
ニュース	BBC News - Home	Chelsea fans urged to stop Abramovich chants	https://www.bbc.co.uk/sport/football/60704463?at_medium=RSS&at_campaign=KARANGA	abramovich	2022-03-11 12:23:32
ニュース	BBC News - Home	England 'really optimistic' unwell Itoje will be fit to face Ireland on Saturday	https://www.bbc.co.uk/sport/rugby-union/60667484?at_medium=RSS&at_campaign=KARANGA	England x really optimistic x unwell Itoje will be fit to face Ireland on SaturdayEngland lock Maro Itoje was unwell overnight but scrum coach Matt Proudfoot is really optimistic he will be able to play in Saturday s Six Nations match against Ireland	2022-03-11 12:31:22
ニュース	BBC News - Home	MOTD Rewind: Man Utd 1-6 Tottenham - How Spurs 'embarrassed' Manchester United in Old Trafford rout	https://www.bbc.co.uk/sport/av/football/58903862?at_medium=RSS&at_campaign=KARANGA	MOTD Rewind Man Utd Tottenham How Spurs x embarrassed x Manchester United in Old Trafford routWatch how Tottenham Hotspur pulled off one of the shock results of the season with a win over Manchester United at Old Trafford	2022-03-11 12:56:06
北海道	北海道新聞	ＪＲ西８００人削減へ　定年、採用抑制でスリム化	https://www.hokkaido-np.co.jp/article/655916/	採用抑制	2022-03-11 21:15:00
北海道	北海道新聞	国の不作為指摘、全面救済迫る　強制不妊訴訟・東京高裁判決	https://www.hokkaido-np.co.jp/article/655915/	強制不妊	2022-03-11 21:13:00
北海道	北海道新聞	桜の通り抜け３年ぶり実施、大阪　造幣局、春の風物詩	https://www.hokkaido-np.co.jp/article/655914/	春の風物詩	2022-03-11 21:12:00
北海道	北海道新聞	選手と寮暮らし、日誌は「生きる道の点検」　クラーク高野球部佐々木監督	https://www.hokkaido-np.co.jp/article/655913/	佐々木啓司	2022-03-11 21:11:00
北海道	北海道新聞	ウクライナ侵攻、食品流通に打撃　ロシア上空の空路使えず　サーモン入荷滞り、小麦価格も上昇見込み	https://www.hokkaido-np.co.jp/article/655912/	食品	2022-03-11 21:08:00
IT	週刊アスキー	ガンダムチームシューター『GUNDAM EVOLUTION』ゲーム本編の全BGMを牧野忠義氏（スピンソルファ）が担当！	https://weekly.ascii.jp/elem/000/004/085/4085958/	gundamevolution	2022-03-11 21:35:00
IT	週刊アスキー	『LOST ARK』の最新アップデートを紹介する生放送「Pmangのゲムづめ！#96」が3月15日に配信決定！	https://weekly.ascii.jp/elem/000/004/085/4085957/	lostark	2022-03-11 21:10:00